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/PowerPC/resolvefi-basereg.ll

363 lines
24 KiB
LLVM
Raw Normal View History

[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
; RUN: llc -O0 -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 < %s | FileCheck %s
; Due to a bug in resolveFrameIndex we ended up with invalid addresses
; containing a base register 0. Verify that this no longer happens.
; CHECK-NOT: (0)
target datalayout = "E-m:e-i64:64-n32:64"
target triple = "powerpc64-unknown-linux-gnu"
%struct.Info = type { i32, i32, i8*, i8*, i8*, [32 x i8*], i64, [32 x i64], i64, i64, i64, [32 x i64] }
%struct.S1998 = type { [2 x i32*], i64, i64, double, i16, i32, [29 x %struct.anon], i16, i8, i32, [8 x i8] }
%struct.anon = type { [16 x double], i32, i16, i32, [3 x i8], [6 x i8], [4 x i32], i8 }
@info = global %struct.Info zeroinitializer, align 8
@fails = global i32 0, align 4
@intarray = global [256 x i32] zeroinitializer, align 4
@s1998 = global %struct.S1998 zeroinitializer, align 16
@a1998 = external global [5 x %struct.S1998]
define void @test1998() {
entry:
%i = alloca i32, align 4
%j = alloca i32, align 4
%tmp = alloca i32, align 4
%agg.tmp = alloca %struct.S1998, align 16
%agg.tmp111 = alloca %struct.S1998, align 16
%agg.tmp112 = alloca %struct.S1998, align 16
%agg.tmp113 = alloca %struct.S1998, align 16
%agg.tmp114 = alloca %struct.S1998, align 16
%agg.tmp115 = alloca %struct.S1998, align 16
%agg.tmp116 = alloca %struct.S1998, align 16
%agg.tmp117 = alloca %struct.S1998, align 16
%agg.tmp118 = alloca %struct.S1998, align 16
%agg.tmp119 = alloca %struct.S1998, align 16
call void @llvm.memset.p0i8.i64(i8* bitcast (%struct.S1998* @s1998 to i8*), i8 0, i64 5168, i32 16, i1 false)
call void @llvm.memset.p0i8.i64(i8* bitcast ([5 x %struct.S1998]* @a1998 to i8*), i8 0, i64 25840, i32 16, i1 false)
call void @llvm.memset.p0i8.i64(i8* bitcast (%struct.Info* @info to i8*), i8 0, i64 832, i32 8, i1 false)
store i8* bitcast (%struct.S1998* @s1998 to i8*), i8** getelementptr inbounds (%struct.Info* @info, i32 0, i32 2), align 8
store i8* bitcast ([5 x %struct.S1998]* @a1998 to i8*), i8** getelementptr inbounds (%struct.Info* @info, i32 0, i32 3), align 8
store i8* bitcast (%struct.S1998* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 3) to i8*), i8** getelementptr inbounds (%struct.Info* @info, i32 0, i32 4), align 8
store i64 5168, i64* getelementptr inbounds (%struct.Info* @info, i32 0, i32 6), align 8
store i64 16, i64* getelementptr inbounds (%struct.Info* @info, i32 0, i32 8), align 8
store i64 16, i64* getelementptr inbounds (%struct.Info* @info, i32 0, i32 9), align 8
store i64 16, i64* getelementptr inbounds (%struct.Info* @info, i32 0, i32 10), align 8
%0 = load i64* getelementptr inbounds (%struct.Info* @info, i32 0, i32 8), align 8
%sub = sub i64 %0, 1
%and = and i64 ptrtoint (%struct.S1998* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 3) to i64), %sub
%tobool = icmp ne i64 %and, 0
br i1 %tobool, label %if.then, label %if.end
if.then: ; preds = %entry
%1 = load i32* @fails, align 4
%inc = add nsw i32 %1, 1
store i32 %inc, i32* @fails, align 4
br label %if.end
if.end: ; preds = %if.then, %entry
store i32 0, i32* %i, align 4
store i32 0, i32* %j, align 4
%2 = load i32* %i, align 4
%idxprom = sext i32 %2 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (i32** getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 0, i64 1) to i8*), i8** %arrayidx, align 8
%3 = load i32* %i, align 4
%idxprom1 = sext i32 %3 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx2 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom1
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx2, align 8
%4 = load i32* %i, align 4
%idxprom3 = sext i32 %4 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx4 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom3
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx4, align 8
store i32* getelementptr inbounds ([256 x i32]* @intarray, i32 0, i64 190), i32** getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 0, i64 1), align 8
store i32* getelementptr inbounds ([256 x i32]* @intarray, i32 0, i64 241), i32** getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 0, i64 1), align 8
%5 = load i32* %i, align 4
%inc5 = add nsw i32 %5, 1
store i32 %inc5, i32* %i, align 4
%6 = load i32* %i, align 4
%idxprom6 = sext i32 %6 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx7 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom6
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (i64* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 1) to i8*), i8** %arrayidx7, align 8
%7 = load i32* %i, align 4
%idxprom8 = sext i32 %7 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx9 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom8
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx9, align 8
%8 = load i32* %i, align 4
%idxprom10 = sext i32 %8 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx11 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom10
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx11, align 8
store i64 -3866974208859106459, i64* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 1), align 8
store i64 -185376695371304091, i64* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 1), align 8
%9 = load i32* %i, align 4
%inc12 = add nsw i32 %9, 1
store i32 %inc12, i32* %i, align 4
%10 = load i32* %i, align 4
%idxprom13 = sext i32 %10 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx14 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom13
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (i64* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 2) to i8*), i8** %arrayidx14, align 8
%11 = load i32* %i, align 4
%idxprom15 = sext i32 %11 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx16 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom15
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx16, align 8
%12 = load i32* %i, align 4
%idxprom17 = sext i32 %12 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx18 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom17
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx18, align 8
store i64 -963638028680427187, i64* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 2), align 8
store i64 7510542175772455554, i64* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 2), align 8
%13 = load i32* %i, align 4
%inc19 = add nsw i32 %13, 1
store i32 %inc19, i32* %i, align 4
%14 = load i32* %i, align 4
%idxprom20 = sext i32 %14 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx21 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom20
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (double* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 3) to i8*), i8** %arrayidx21, align 8
%15 = load i32* %i, align 4
%idxprom22 = sext i32 %15 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx23 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom22
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx23, align 8
%16 = load i32* %i, align 4
%idxprom24 = sext i32 %16 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx25 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom24
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 16, i64* %arrayidx25, align 8
store double 0xC0F8783300000000, double* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 3), align 16
store double 0xC10DF3CCC0000000, double* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 3), align 16
%17 = load i32* %i, align 4
%inc26 = add nsw i32 %17, 1
store i32 %inc26, i32* %i, align 4
%18 = load i32* %i, align 4
%idxprom27 = sext i32 %18 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx28 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom27
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (i16* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 4) to i8*), i8** %arrayidx28, align 8
%19 = load i32* %i, align 4
%idxprom29 = sext i32 %19 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx30 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom29
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 2, i64* %arrayidx30, align 8
%20 = load i32* %i, align 4
%idxprom31 = sext i32 %20 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx32 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom31
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 2, i64* %arrayidx32, align 8
store i16 -15897, i16* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 4), align 2
store i16 30935, i16* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 4), align 2
%21 = load i32* %i, align 4
%inc33 = add nsw i32 %21, 1
store i32 %inc33, i32* %i, align 4
store i32 -419541644, i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 5), align 4
store i32 2125926812, i32* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 5), align 4
%22 = load i32* %j, align 4
%inc34 = add nsw i32 %22, 1
store i32 %inc34, i32* %j, align 4
%23 = load i32* %i, align 4
%idxprom35 = sext i32 %23 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx36 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom35
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (double* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 0, i64 0) to i8*), i8** %arrayidx36, align 8
%24 = load i32* %i, align 4
%idxprom37 = sext i32 %24 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx38 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom37
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx38, align 8
%25 = load i32* %i, align 4
%idxprom39 = sext i32 %25 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx40 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom39
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 8, i64* %arrayidx40, align 8
store double 0xC0FC765780000000, double* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 0, i64 0), align 8
store double 0xC1025CD7A0000000, double* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 0, i64 0), align 8
%26 = load i32* %i, align 4
%inc41 = add nsw i32 %26, 1
store i32 %inc41, i32* %i, align 4
%bf.load = load i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 1), align 8
%bf.clear = and i32 %bf.load, 7
%bf.set = or i32 %bf.clear, 16
store i32 %bf.set, i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 1), align 8
%bf.load42 = load i32* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 1), align 8
%bf.clear43 = and i32 %bf.load42, 7
%bf.set44 = or i32 %bf.clear43, 24
store i32 %bf.set44, i32* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 1), align 8
%27 = load i32* %j, align 4
%inc45 = add nsw i32 %27, 1
store i32 %inc45, i32* %j, align 4
%bf.load46 = load i16* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 2), align 4
%bf.clear47 = and i16 %bf.load46, 127
store i16 %bf.clear47, i16* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 2), align 4
%bf.load48 = load i16* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 2), align 4
%bf.clear49 = and i16 %bf.load48, 127
store i16 %bf.clear49, i16* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 2), align 4
%28 = load i32* %j, align 4
%inc50 = add nsw i32 %28, 1
store i32 %inc50, i32* %j, align 4
%bf.load51 = load i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 3), align 8
%bf.clear52 = and i32 %bf.load51, 63
store i32 %bf.clear52, i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 3), align 8
%bf.load53 = load i32* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 3), align 8
%bf.clear54 = and i32 %bf.load53, 63
%bf.set55 = or i32 %bf.clear54, 64
store i32 %bf.set55, i32* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 3), align 8
%29 = load i32* %j, align 4
%inc56 = add nsw i32 %29, 1
store i32 %inc56, i32* %j, align 4
%bf.load57 = load i24* bitcast ([3 x i8]* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 4) to i24*), align 4
%bf.clear58 = and i24 %bf.load57, 63
store i24 %bf.clear58, i24* bitcast ([3 x i8]* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 4) to i24*), align 4
%bf.load59 = load i24* bitcast ([3 x i8]* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 4) to i24*), align 4
%bf.clear60 = and i24 %bf.load59, 63
store i24 %bf.clear60, i24* bitcast ([3 x i8]* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 4) to i24*), align 4
%30 = load i32* %j, align 4
%inc61 = add nsw i32 %30, 1
store i32 %inc61, i32* %j, align 4
%31 = load i32* %i, align 4
%idxprom62 = sext i32 %31 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx63 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom62
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 5, i64 5), i8** %arrayidx63, align 8
%32 = load i32* %i, align 4
%idxprom64 = sext i32 %32 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx65 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom64
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx65, align 8
%33 = load i32* %i, align 4
%idxprom66 = sext i32 %33 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx67 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom66
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx67, align 8
store i8 -83, i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 5, i64 5), align 1
store i8 -67, i8* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 5, i64 5), align 1
%34 = load i32* %i, align 4
%inc68 = add nsw i32 %34, 1
store i32 %inc68, i32* %i, align 4
%35 = load i32* %i, align 4
%idxprom69 = sext i32 %35 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx70 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom69
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 5, i64 1), i8** %arrayidx70, align 8
%36 = load i32* %i, align 4
%idxprom71 = sext i32 %36 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx72 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom71
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx72, align 8
%37 = load i32* %i, align 4
%idxprom73 = sext i32 %37 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx74 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom73
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx74, align 8
store i8 34, i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 5, i64 1), align 1
store i8 64, i8* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 5, i64 1), align 1
%38 = load i32* %i, align 4
%inc75 = add nsw i32 %38, 1
store i32 %inc75, i32* %i, align 4
%39 = load i32* %i, align 4
%idxprom76 = sext i32 %39 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx77 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom76
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 6, i64 3) to i8*), i8** %arrayidx77, align 8
%40 = load i32* %i, align 4
%idxprom78 = sext i32 %40 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx79 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom78
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 4, i64* %arrayidx79, align 8
%41 = load i32* %i, align 4
%idxprom80 = sext i32 %41 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx81 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom80
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 4, i64* %arrayidx81, align 8
store i32 -3, i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 6, i64 3), align 4
store i32 -3, i32* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 6, i64 3), align 4
%42 = load i32* %i, align 4
%inc82 = add nsw i32 %42, 1
store i32 %inc82, i32* %i, align 4
%43 = load i32* %i, align 4
%idxprom83 = sext i32 %43 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx84 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom83
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 7), i8** %arrayidx84, align 8
%44 = load i32* %i, align 4
%idxprom85 = sext i32 %44 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx86 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom85
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx86, align 8
%45 = load i32* %i, align 4
%idxprom87 = sext i32 %45 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx88 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom87
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx88, align 8
store i8 106, i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 6, i64 4, i32 7), align 1
store i8 -102, i8* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 6, i64 4, i32 7), align 1
%46 = load i32* %i, align 4
%inc89 = add nsw i32 %46, 1
store i32 %inc89, i32* %i, align 4
%47 = load i32* %i, align 4
%idxprom90 = sext i32 %47 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx91 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom90
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (i16* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 7) to i8*), i8** %arrayidx91, align 8
%48 = load i32* %i, align 4
%idxprom92 = sext i32 %48 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx93 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom92
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 2, i64* %arrayidx93, align 8
%49 = load i32* %i, align 4
%idxprom94 = sext i32 %49 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx95 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom94
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 2, i64* %arrayidx95, align 8
store i16 29665, i16* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 7), align 2
store i16 7107, i16* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 7), align 2
%50 = load i32* %i, align 4
%inc96 = add nsw i32 %50, 1
store i32 %inc96, i32* %i, align 4
%51 = load i32* %i, align 4
%idxprom97 = sext i32 %51 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx98 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom97
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 8), i8** %arrayidx98, align 8
%52 = load i32* %i, align 4
%idxprom99 = sext i32 %52 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx100 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom99
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx100, align 8
%53 = load i32* %i, align 4
%idxprom101 = sext i32 %53 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx102 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom101
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 1, i64* %arrayidx102, align 8
store i8 52, i8* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 8), align 1
store i8 -86, i8* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 8), align 1
%54 = load i32* %i, align 4
%inc103 = add nsw i32 %54, 1
store i32 %inc103, i32* %i, align 4
%55 = load i32* %i, align 4
%idxprom104 = sext i32 %55 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx105 = getelementptr inbounds [32 x i8*], [32 x i8*]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 5), i32 0, i64 %idxprom104
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i8* bitcast (i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 9) to i8*), i8** %arrayidx105, align 8
%56 = load i32* %i, align 4
%idxprom106 = sext i32 %56 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx107 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 7), i32 0, i64 %idxprom106
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 4, i64* %arrayidx107, align 8
%57 = load i32* %i, align 4
%idxprom108 = sext i32 %57 to i64
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%arrayidx109 = getelementptr inbounds [32 x i64], [32 x i64]* getelementptr inbounds (%struct.Info* @info, i32 0, i32 11), i32 0, i64 %idxprom108
[PowerPC] Constrain base register in PPCRegisterInfo::resolveFrameIndex I've run into a bug where current LLVM at -O0 (with fast-isel) generated invalid code like: ld 0, 20936(1) # 8-byte Folded Reload stw 12, 10348(0) stw 12, 10344(0) The underlying vreg had been introduced as base register by the Local Stack Slot Allocation pass. That register was constrained to G8RC by PPCRegisterInfo::materializeFrameBaseRegister to match the ADDI instruction used to set it, but it was *not* constrained to G8RC_NOX0 to fit the *use* of the register in an address. That should have happened in PPCRegisterInfo::resolveFrameIndex. This patch adds an appropriate constrainRegClass call. Reviewed by Hal Finkel. llvm-svn: 211897
2014-06-27 21:04:12 +08:00
store i64 4, i64* %arrayidx109, align 8
store i32 -54118453, i32* getelementptr inbounds (%struct.S1998* @s1998, i32 0, i32 9), align 4
store i32 1668755823, i32* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2, i32 9), align 4
%58 = load i32* %i, align 4
%inc110 = add nsw i32 %58, 1
store i32 %inc110, i32* %i, align 4
store i32 %inc110, i32* %tmp
%59 = load i32* %tmp
%60 = load i32* %i, align 4
store i32 %60, i32* getelementptr inbounds (%struct.Info* @info, i32 0, i32 0), align 4
%61 = load i32* %j, align 4
store i32 %61, i32* getelementptr inbounds (%struct.Info* @info, i32 0, i32 1), align 4
%62 = bitcast %struct.S1998* %agg.tmp111 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %62, i8* bitcast (%struct.S1998* @s1998 to i8*), i64 5168, i32 16, i1 false)
%63 = bitcast %struct.S1998* %agg.tmp112 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %63, i8* bitcast (%struct.S1998* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2) to i8*), i64 5168, i32 16, i1 false)
call void @check1998(%struct.S1998* sret %agg.tmp, %struct.S1998* byval align 16 %agg.tmp111, %struct.S1998* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 1), %struct.S1998* byval align 16 %agg.tmp112)
call void @checkx1998(%struct.S1998* byval align 16 %agg.tmp)
%64 = bitcast %struct.S1998* %agg.tmp113 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %64, i8* bitcast (%struct.S1998* @s1998 to i8*), i64 5168, i32 16, i1 false)
%65 = bitcast %struct.S1998* %agg.tmp114 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %65, i8* bitcast (%struct.S1998* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2) to i8*), i64 5168, i32 16, i1 false)
%66 = bitcast %struct.S1998* %agg.tmp115 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %66, i8* bitcast (%struct.S1998* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2) to i8*), i64 5168, i32 16, i1 false)
call void (i32, ...)* @check1998va(i32 signext 1, double 1.000000e+00, %struct.S1998* byval align 16 %agg.tmp113, i64 2, %struct.S1998* byval align 16 %agg.tmp114, %struct.S1998* byval align 16 %agg.tmp115)
%67 = bitcast %struct.S1998* %agg.tmp116 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %67, i8* bitcast (%struct.S1998* @s1998 to i8*), i64 5168, i32 16, i1 false)
%68 = bitcast %struct.S1998* %agg.tmp117 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %68, i8* bitcast (%struct.S1998* @s1998 to i8*), i64 5168, i32 16, i1 false)
%69 = bitcast %struct.S1998* %agg.tmp118 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %69, i8* bitcast (%struct.S1998* getelementptr inbounds ([5 x %struct.S1998]* @a1998, i32 0, i64 2) to i8*), i64 5168, i32 16, i1 false)
%70 = bitcast %struct.S1998* %agg.tmp119 to i8*
call void @llvm.memcpy.p0i8.p0i8.i64(i8* %70, i8* bitcast (%struct.S1998* @s1998 to i8*), i64 5168, i32 16, i1 false)
call void (i32, ...)* @check1998va(i32 signext 2, %struct.S1998* byval align 16 %agg.tmp116, %struct.S1998* byval align 16 %agg.tmp117, ppc_fp128 0xM40000000000000000000000000000000, %struct.S1998* byval align 16 %agg.tmp118, %struct.S1998* byval align 16 %agg.tmp119)
ret void
}
declare void @llvm.memset.p0i8.i64(i8* nocapture, i8, i64, i32, i1)
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture, i8* nocapture readonly, i64, i32, i1)
declare void @check1998(%struct.S1998* sret, %struct.S1998* byval align 16, %struct.S1998*, %struct.S1998* byval align 16)
declare void @check1998va(i32 signext, ...)
declare void @checkx1998(%struct.S1998* byval align 16 %arg)