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/Mips/msa/frameindex.ll

395 lines
13 KiB
LLVM
Raw Normal View History

[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=MIPS32-AE -check-prefix=MIPS32-BE %s
; RUN: llc -march=mipsel -mattr=+msa,+fp64 < %s | FileCheck -check-prefix=MIPS32-AE -check-prefix=MIPS32-LE %s
define void @loadstore_v16i8_near() nounwind {
; MIPS32-AE: loadstore_v16i8_near:
%1 = alloca <16 x i8>
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%2 = load volatile <16 x i8>, <16 x i8>* %1
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: ld.b [[R1:\$w[0-9]+]], 0($sp)
store volatile <16 x i8> %2, <16 x i8>* %1
; MIPS32-AE: st.b [[R1]], 0($sp)
ret void
; MIPS32-AE: .size loadstore_v16i8_near
}
define void @loadstore_v16i8_just_under_simm10() nounwind {
; MIPS32-AE: loadstore_v16i8_just_under_simm10:
%1 = alloca <16 x i8>
%2 = alloca [496 x i8] ; Push the frame right up to 512 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <16 x i8>, <16 x i8>* %1
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: ld.b [[R1:\$w[0-9]+]], 496($sp)
store volatile <16 x i8> %3, <16 x i8>* %1
; MIPS32-AE: st.b [[R1]], 496($sp)
ret void
; MIPS32-AE: .size loadstore_v16i8_just_under_simm10
}
define void @loadstore_v16i8_just_over_simm10() nounwind {
; MIPS32-AE: loadstore_v16i8_just_over_simm10:
%1 = alloca <16 x i8>
%2 = alloca [497 x i8] ; Push the frame just over 512 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <16 x i8>, <16 x i8>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 512
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: ld.b [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <16 x i8> %3, <16 x i8>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 512
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: st.b [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v16i8_just_over_simm10
}
define void @loadstore_v16i8_just_under_simm16() nounwind {
; MIPS32-AE: loadstore_v16i8_just_under_simm16:
%1 = alloca <16 x i8>
%2 = alloca [32752 x i8] ; Push the frame right up to 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <16 x i8>, <16 x i8>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: ld.b [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <16 x i8> %3, <16 x i8>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: st.b [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v16i8_just_under_simm16
}
define void @loadstore_v16i8_just_over_simm16() nounwind {
; MIPS32-AE: loadstore_v16i8_just_over_simm16:
%1 = alloca <16 x i8>
%2 = alloca [32753 x i8] ; Push the frame just over 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <16 x i8>, <16 x i8>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: ld.b [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <16 x i8> %3, <16 x i8>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] MSA loads and stores have a 10-bit offset. Account for this when lowering FrameIndex. This prevents the compiler from emitting invalid ld.[bhwd]'s and st.[bhwd]'s when the stack frame is between 512 and 32,768 bytes in size. llvm-svn: 195973
2013-11-30 21:47:57 +08:00
; MIPS32-AE: st.b [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v16i8_just_over_simm16
}
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
define void @loadstore_v8i16_near() nounwind {
; MIPS32-AE: loadstore_v8i16_near:
%1 = alloca <8 x i16>
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%2 = load volatile <8 x i16>, <8 x i16>* %1
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.h [[R1:\$w[0-9]+]], 0($sp)
store volatile <8 x i16> %2, <8 x i16>* %1
; MIPS32-AE: st.h [[R1]], 0($sp)
ret void
; MIPS32-AE: .size loadstore_v8i16_near
}
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
define void @loadstore_v8i16_unaligned() nounwind {
; MIPS32-AE: loadstore_v8i16_unaligned:
%1 = alloca [2 x <8 x i16>]
%2 = bitcast [2 x <8 x i16>]* %1 to i8*
[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
%3 = getelementptr i8, i8* %2, i32 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
%4 = bitcast i8* %3 to [2 x <8 x i16>]*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%5 = getelementptr [2 x <8 x i16>], [2 x <8 x i16>]* %4, i32 0, i32 0
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%6 = load volatile <8 x i16>, <8 x i16>* %5
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
; MIPS32-AE: ld.h [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <8 x i16> %6, <8 x i16>* %5
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
; MIPS32-AE: st.h [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v8i16_unaligned
}
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
define void @loadstore_v8i16_just_under_simm10() nounwind {
; MIPS32-AE: loadstore_v8i16_just_under_simm10:
%1 = alloca <8 x i16>
%2 = alloca [1008 x i8] ; Push the frame right up to 1024 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <8 x i16>, <8 x i16>* %1
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.h [[R1:\$w[0-9]+]], 1008($sp)
store volatile <8 x i16> %3, <8 x i16>* %1
; MIPS32-AE: st.h [[R1]], 1008($sp)
ret void
; MIPS32-AE: .size loadstore_v8i16_just_under_simm10
}
define void @loadstore_v8i16_just_over_simm10() nounwind {
; MIPS32-AE: loadstore_v8i16_just_over_simm10:
%1 = alloca <8 x i16>
%2 = alloca [1009 x i8] ; Push the frame just over 1024 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <8 x i16>, <8 x i16>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1024
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.h [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <8 x i16> %3, <8 x i16>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1024
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.h [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v8i16_just_over_simm10
}
define void @loadstore_v8i16_just_under_simm16() nounwind {
; MIPS32-AE: loadstore_v8i16_just_under_simm16:
%1 = alloca <8 x i16>
%2 = alloca [32752 x i8] ; Push the frame right up to 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <8 x i16>, <8 x i16>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.h [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <8 x i16> %3, <8 x i16>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.h [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v8i16_just_under_simm16
}
define void @loadstore_v8i16_just_over_simm16() nounwind {
; MIPS32-AE: loadstore_v8i16_just_over_simm16:
%1 = alloca <8 x i16>
%2 = alloca [32753 x i8] ; Push the frame just over 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <8 x i16>, <8 x i16>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.h [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <8 x i16> %3, <8 x i16>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.h [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v8i16_just_over_simm16
}
define void @loadstore_v4i32_near() nounwind {
; MIPS32-AE: loadstore_v4i32_near:
%1 = alloca <4 x i32>
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%2 = load volatile <4 x i32>, <4 x i32>* %1
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.w [[R1:\$w[0-9]+]], 0($sp)
store volatile <4 x i32> %2, <4 x i32>* %1
; MIPS32-AE: st.w [[R1]], 0($sp)
ret void
; MIPS32-AE: .size loadstore_v4i32_near
}
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
define void @loadstore_v4i32_unaligned() nounwind {
; MIPS32-AE: loadstore_v4i32_unaligned:
%1 = alloca [2 x <4 x i32>]
%2 = bitcast [2 x <4 x i32>]* %1 to i8*
[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
%3 = getelementptr i8, i8* %2, i32 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
%4 = bitcast i8* %3 to [2 x <4 x i32>]*
[opaque pointer type] Add textual IR support for explicit type parameter to getelementptr instruction One of several parallel first steps to remove the target type of pointers, replacing them with a single opaque pointer type. This adds an explicit type parameter to the gep instruction so that when the first parameter becomes an opaque pointer type, the type to gep through is still available to the instructions. * This doesn't modify gep operators, only instructions (operators will be handled separately) * Textual IR changes only. Bitcode (including upgrade) and changing the in-memory representation will be in separate changes. * geps of vectors are transformed as: getelementptr <4 x float*> %x, ... ->getelementptr float, <4 x float*> %x, ... Then, once the opaque pointer type is introduced, this will ultimately look like: getelementptr float, <4 x ptr> %x with the unambiguous interpretation that it is a vector of pointers to float. * address spaces remain on the pointer, not the type: getelementptr float addrspace(1)* %x ->getelementptr float, float addrspace(1)* %x Then, eventually: getelementptr float, ptr addrspace(1) %x Importantly, the massive amount of test case churn has been automated by same crappy python code. I had to manually update a few test cases that wouldn't fit the script's model (r228970,r229196,r229197,r229198). The python script just massages stdin and writes the result to stdout, I then wrapped that in a shell script to handle replacing files, then using the usual find+xargs to migrate all the files. update.py: import fileinput import sys import re ibrep = re.compile(r"(^.*?[^%\w]getelementptr inbounds )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") normrep = re.compile( r"(^.*?[^%\w]getelementptr )(((?:<\d* x )?)(.*?)(| addrspace\(\d\)) *\*(|>)(?:$| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$))") def conv(match, line): if not match: return line line = match.groups()[0] if len(match.groups()[5]) == 0: line += match.groups()[2] line += match.groups()[3] line += ", " line += match.groups()[1] line += "\n" return line for line in sys.stdin: if line.find("getelementptr ") == line.find("getelementptr inbounds"): if line.find("getelementptr inbounds") != line.find("getelementptr inbounds ("): line = conv(re.match(ibrep, line), line) elif line.find("getelementptr ") != line.find("getelementptr ("): line = conv(re.match(normrep, line), line) sys.stdout.write(line) apply.sh: for name in "$@" do python3 `dirname "$0"`/update.py < "$name" > "$name.tmp" && mv "$name.tmp" "$name" rm -f "$name.tmp" done The actual commands: From llvm/src: find test/ -name *.ll | xargs ./apply.sh From llvm/src/tools/clang: find test/ -name *.mm -o -name *.m -o -name *.cpp -o -name *.c | xargs -I '{}' ../../apply.sh "{}" From llvm/src/tools/polly: find test/ -name *.ll | xargs ./apply.sh After that, check-all (with llvm, clang, clang-tools-extra, lld, compiler-rt, and polly all checked out). The extra 'rm' in the apply.sh script is due to a few files in clang's test suite using interesting unicode stuff that my python script was throwing exceptions on. None of those files needed to be migrated, so it seemed sufficient to ignore those cases. Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7636 llvm-svn: 230786
2015-02-28 03:29:02 +08:00
%5 = getelementptr [2 x <4 x i32>], [2 x <4 x i32>]* %4, i32 0, i32 0
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%6 = load volatile <4 x i32>, <4 x i32>* %5
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
; MIPS32-AE: ld.w [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <4 x i32> %6, <4 x i32>* %5
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
; MIPS32-AE: st.w [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v4i32_unaligned
}
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
define void @loadstore_v4i32_just_under_simm10() nounwind {
; MIPS32-AE: loadstore_v4i32_just_under_simm10:
%1 = alloca <4 x i32>
%2 = alloca [2032 x i8] ; Push the frame right up to 2048 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <4 x i32>, <4 x i32>* %1
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.w [[R1:\$w[0-9]+]], 2032($sp)
store volatile <4 x i32> %3, <4 x i32>* %1
; MIPS32-AE: st.w [[R1]], 2032($sp)
ret void
; MIPS32-AE: .size loadstore_v4i32_just_under_simm10
}
define void @loadstore_v4i32_just_over_simm10() nounwind {
; MIPS32-AE: loadstore_v4i32_just_over_simm10:
%1 = alloca <4 x i32>
%2 = alloca [2033 x i8] ; Push the frame just over 2048 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <4 x i32>, <4 x i32>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 2048
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.w [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <4 x i32> %3, <4 x i32>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 2048
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.w [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v4i32_just_over_simm10
}
define void @loadstore_v4i32_just_under_simm16() nounwind {
; MIPS32-AE: loadstore_v4i32_just_under_simm16:
%1 = alloca <4 x i32>
%2 = alloca [32752 x i8] ; Push the frame right up to 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <4 x i32>, <4 x i32>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.w [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <4 x i32> %3, <4 x i32>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.w [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v4i32_just_under_simm16
}
define void @loadstore_v4i32_just_over_simm16() nounwind {
; MIPS32-AE: loadstore_v4i32_just_over_simm16:
%1 = alloca <4 x i32>
%2 = alloca [32753 x i8] ; Push the frame just over 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <4 x i32>, <4 x i32>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.w [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <4 x i32> %3, <4 x i32>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.w [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v4i32_just_over_simm16
}
define void @loadstore_v2i64_near() nounwind {
; MIPS32-AE: loadstore_v2i64_near:
%1 = alloca <2 x i64>
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%2 = load volatile <2 x i64>, <2 x i64>* %1
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.d [[R1:\$w[0-9]+]], 0($sp)
store volatile <2 x i64> %2, <2 x i64>* %1
; MIPS32-AE: st.d [[R1]], 0($sp)
ret void
; MIPS32-AE: .size loadstore_v2i64_near
}
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
define void @loadstore_v2i64_unaligned() nounwind {
; MIPS32-AE: loadstore_v2i64_unaligned:
%1 = alloca [2 x <2 x i64>]
%2 = bitcast [2 x <2 x i64>]* %1 to i8*
[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
%3 = getelementptr i8, i8* %2, i32 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
%4 = bitcast i8* %3 to [2 x <2 x 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
%5 = getelementptr [2 x <2 x i64>], [2 x <2 x i64>]* %4, i32 0, i32 0
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%6 = load volatile <2 x i64>, <2 x i64>* %5
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
; MIPS32-AE: ld.d [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <2 x i64> %6, <2 x i64>* %5
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 1
[mips][msa] Fix invalid generated code when lowering FrameIndex involving unaligned offsets. Summary: The MSA ld.[bhwd] and st.[bhwd] instructions scale the immediate by the element size before use as an offset. The offset must therefore be a multiple of the element size to be valid in these instructions. However, an unaligned base address is valid in MSA. This commit causes the compiler to emit valid code when the calculated offset is not a multiple of the element size by accounting for the offset using addiu and using a zero offset in the load/store. Depends on D2338 Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2339 llvm-svn: 196777
2013-12-09 20:47:12 +08:00
; MIPS32-AE: st.d [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v2i64_unaligned
}
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
define void @loadstore_v2i64_just_under_simm10() nounwind {
; MIPS32-AE: loadstore_v2i64_just_under_simm10:
%1 = alloca <2 x i64>
%2 = alloca [4080 x i8] ; Push the frame right up to 4096 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <2 x i64>, <2 x i64>* %1
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.d [[R1:\$w[0-9]+]], 4080($sp)
store volatile <2 x i64> %3, <2 x i64>* %1
; MIPS32-AE: st.d [[R1]], 4080($sp)
ret void
; MIPS32-AE: .size loadstore_v2i64_just_under_simm10
}
define void @loadstore_v2i64_just_over_simm10() nounwind {
; MIPS32-AE: loadstore_v2i64_just_over_simm10:
%1 = alloca <2 x i64>
%2 = alloca [4081 x i8] ; Push the frame just over 4096 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <2 x i64>, <2 x i64>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 4096
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.d [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <2 x i64> %3, <2 x i64>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: addiu [[BASE:\$([0-9]+|gp)]], $sp, 4096
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.d [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v2i64_just_over_simm10
}
define void @loadstore_v2i64_just_under_simm16() nounwind {
; MIPS32-AE: loadstore_v2i64_just_under_simm16:
%1 = alloca <2 x i64>
%2 = alloca [32752 x i8] ; Push the frame right up to 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <2 x i64>, <2 x i64>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.d [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <2 x i64> %3, <2 x i64>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.d [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v2i64_just_under_simm16
}
define void @loadstore_v2i64_just_over_simm16() nounwind {
; MIPS32-AE: loadstore_v2i64_just_over_simm16:
%1 = alloca <2 x i64>
%2 = alloca [32753 x i8] ; Push the frame just over 32768 bytes
[opaque pointer type] Add textual IR support for explicit type parameter to load instruction Essentially the same as the GEP change in r230786. A similar migration script can be used to update test cases, though a few more test case improvements/changes were required this time around: (r229269-r229278) import fileinput import sys import re pat = re.compile(r"((?:=|:|^)\s*load (?:atomic )?(?:volatile )?(.*?))(| addrspace\(\d+\) *)\*($| *(?:%|@|null|undef|blockaddress|getelementptr|addrspacecast|bitcast|inttoptr|\[\[[a-zA-Z]|\{\{).*$)") for line in sys.stdin: sys.stdout.write(re.sub(pat, r"\1, \2\3*\4", line)) Reviewers: rafael, dexonsmith, grosser Differential Revision: http://reviews.llvm.org/D7649 llvm-svn: 230794
2015-02-28 05:17:42 +08:00
%3 = load volatile <2 x i64>, <2 x i64>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: ld.d [[R1:\$w[0-9]+]], 0([[BASE]])
store volatile <2 x i64> %3, <2 x i64>* %1
[mips] Improve robustness of some tests. Summary: This is done by removing some hardcoded registers like $at or expecting a single digit register to be selected. Contains work done by Matheus Almeida. Reviewers: matheusalmeida, dsanders Reviewed By: dsanders Subscribers: tomatabacu Differential Revision: http://reviews.llvm.org/D4227 llvm-svn: 215640
2014-08-14 21:10:48 +08:00
; MIPS32-AE: ori [[R2:\$([0-9]+|gp)]], $zero, 32768
; MIPS32-AE: addu [[BASE:\$([0-9]+|gp)]], $sp, [[R2]]
[mips][msa] Fix suboptimal FrameIndex lowering for ld.[hwd] and st.[hwd] Summary: The immediate in these instructions is scaled before use as an offset. They therefore have a wider reach than ld.b/st.b. Reviewers: matheusalmeida Reviewed By: matheusalmeida Differential Revision: http://llvm-reviews.chandlerc.com/D2338 llvm-svn: 196775
2013-12-09 19:50:16 +08:00
; MIPS32-AE: st.d [[R1]], 0([[BASE]])
ret void
; MIPS32-AE: .size loadstore_v2i64_just_over_simm16
}