llvmbot
commented
9 years ago Thanks for your quick fix!
Closed llvmbot closed 9 years ago
llvmbot
commented
9 years ago Thanks for your quick fix!
ahmedbougacha
commented
9 years ago Should be fixed with r235243.
ahmedbougacha
commented
9 years ago And here's a from-scratch testcase, with a ctpop to force vector ops legalization to run:
define <4 x i16> @foo(i16* %bar, i16 %ptr, i64 %inc, <4 x i16> %A, <2 x i32> %d1, <2 x i32> %d2) { %tmp1 = load i16, i16 %bar %tmp2 = insertelement <4 x i16> %A, i16 %tmp1, i32 1 %tmp3 = getelementptr i16, i16 %bar, i64 %inc store i16 %tmp3, i16 %ptr %d3 = call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %d1) store <2 x i32> %d3, <2 x i32>* %d2 ret <4 x i16> %tmp2 }
ahmedbougacha
commented
9 years ago Note that this doesn't reproduce on the simple testcases of arm64-indexed-vector-ldst.ll, because there, the v4i16 insert_vector_elt is lowered to extract_subreg of v8i16, and the LD1LANEpost combine doesn't run BeforeLegalizeOps.
isBeforeLegalizeOps is defined as: Level < AfterLegalizeVectorOps
So, the only situation where the combine could run on v4i16 is if there's a vector legalization phase, whereas it should really be able to run after type legalization if there isn't a vector legalization phase.
ahmedbougacha
commented
9 years ago The faulty code is in AArch64DAGToDAGISel::SelectPostLoadLane:
ReplaceUses(SDValue(N, 0),
Narrow ? NarrowVector(SuperReg, *CurDAG) : SuperReg);Here SuperReg is an Untyped vector list (whereas NarrowVector expects a 128-bit vector type), and should probably be wrapped in an EXTRACT_SUBREG.
llvmbot
commented
9 years ago Hi Yao:
Maybe you want to compress filterbanks_neon-c2acf1.c before uploading it (with bzip2 or xz, for example).
Done, thanks!
llvmbot
commented
9 years ago llvmbot
commented
9 years ago Hi Yao:
Maybe you want to compress filterbanks_neon-c2acf1.c before uploading it (with bzip2 or xz, for example).
llvmbot
commented
9 years ago By the way, it is OK when compile with option "-O0".
llvmbot
commented
9 years ago My /tmp/filterbanks_neon-c2acf1.c (3.1MB) file can't be upload due to the size limitation of attachment (1MB).
Extended Description
Clang fails on compile arm neon intrinsic file for aarch64 target.
Here is the log:
/home/zhongwei/projects/read_learn/llvm-toolchain/build/bin/clang -v -target aarch64-none-linux-androideabi -O2 -D__ARM_NEON -c ../../webrtc/modules/audio_coding/codecs/isac/fix/source/filterbanks_neon.c clang version 3.7.0 (trunk 235161) Target: aarch64-none-linux-androideabi Thread model: posix Found candidate GCC installation: /usr/lib/gcc-cross/aarch64-linux-gnu/4.8 Found candidate GCC installation: /usr/lib/gcc-cross/aarch64-linux-gnu/4.8.2 Selected GCC installation: /usr/lib/gcc-cross/aarch64-linux-gnu/4.8 Candidate multilib: .;@m64 Selected multilib: .;@m64 "/home/zhongwei/projects/read_learn/llvm-toolchain/build/bin/clang-3.7" -cc1 -triple aarch64-none-linux-androideabi -emit-obj -disable-free -main-file-name filterbanks_neon.c -mrelocation-model pic -pic-level 1 -mthread-model posix -mdi\ sable-fp-elim -fmath-errno -masm-verbose -mconstructor-aliases -fuse-init-array -target-cpu generic -target-feature +neon -target-abi aapcs -backend-option -aarch64-fix-cortex-a53-835769=1 -v -dwarf-column-info -fno-unique-section-names \ -coverage-file /home/zhongwei/project/webrtc_git/src/out/iOS_Release/filterbanks_neon.c -resource-dir /home/zhongwei/projects/read_learn/llvm-toolchain/build/bin/../lib/clang/3.7.0 -D __ARM_NEON -internal-isystem /usr/local/include -inte\ rnal-isystem /home/zhongwei/projects/read_learn/llvm-toolchain/build/bin/../lib/clang/3.7.0/include -internal-externc-isystem /include -internal-externc-isystem /usr/include -O2 -fdebug-compilation-dir /home/zhongwei/project/webrtc_git/s\ rc/out/iOS_Release -ferror-limit 19 -fmessage-length 118 -mstackrealign -fallow-half-arguments-and-returns -fno-signed-char -fobjc-runtime=gcc -fdiagnostics-show-option -vectorize-loops -vectorize-slp -o filterbanks_neon.o -x c ../../web\ rtc/modules/audio_coding/codecs/isac/fix/source/filterbanks_neon.c clang -cc1 version 3.7.0 based upon LLVM 3.7.0svn default target x86_64-unknown-linux-gnu ignoring nonexistent directory "/include"
include "..." search starts here:
include <...> search starts here:
/usr/local/include /home/zhongwei/projects/read_learn/llvm-toolchain/build/bin/../lib/clang/3.7.0/include /usr/include End of search list. clang-3.7: /home/zhongwei/projects/read_learn/llvm-toolchain/llvm/include/llvm/CodeGen/ValueTypes.h:226: unsigned int llvm::EVT::getVectorNumElements() const: Assertion `isVector() && "Invalid vector type!"' failed. 0 clang-3.7 0x00000000025e7e3a llvm::sys::PrintStackTrace(llvm::raw_ostream&) + 44 1 clang-3.7 0x00000000025e814f 2 clang-3.7 0x00000000025e6c4f 3 libpthread.so.0 0x00002b5d50e4d340 4 libc.so.6 0x00002b5d51acacc9 gsignal + 57 5 libc.so.6 0x00002b5d51ace0d8 abort + 328 6 libc.so.6 0x00002b5d51ac3b86 7 libc.so.6 0x00002b5d51ac3c32 8 clang-3.7 0x0000000001342170 9 clang-3.7 0x000000000134da6e 10 clang-3.7 0x000000000134e4ea 11 clang-3.7 0x000000000135d477 12 clang-3.7 0x0000000002aa7058 llvm::SelectionDAGISel::DoInstructionSelection() + 528 13 clang-3.7 0x0000000002aa69ec llvm::SelectionDAGISel::CodeGenAndEmitDAG() + 3544 14 clang-3.7 0x0000000002aa5845 llvm::SelectionDAGISel::SelectBasicBlock(llvm::ilist_iterator, llvm::ilist_iterator, bool&) + 235
15 clang-3.7 0x0000000002aa8b61 llvm::SelectionDAGISel::SelectAllBasicBlocks(llvm::Function const&) + 2841
16 clang-3.7 0x0000000002aa49d9 llvm::SelectionDAGISel::runOnMachineFunction(llvm::MachineFunction&) + 1013
17 clang-3.7 0x00000000013442a2
18 clang-3.7 0x0000000001e91dcb llvm::MachineFunctionPass::runOnFunction(llvm::Function&) + 95
19 clang-3.7 0x00000000021d8cec llvm::FPPassManager::runOnFunction(llvm::Function&) + 290
19 clang-3.7 0x00000000021d8cec llvm::FPPassManager::runOnFunction(llvm::Function&) + 290
20 clang-3.7 0x00000000021d8e5c llvm::FPPassManager::runOnModule(llvm::Module&) + 84
21 clang-3.7 0x00000000021d917a
22 clang-3.7 0x00000000021d9820 llvm::legacy::PassManagerImpl::run(llvm::Module&) + 244
23 clang-3.7 0x00000000021d9a3f llvm::legacy::PassManager::run(llvm::Module&) + 39
24 clang-3.7 0x0000000002b7083d
25 clang-3.7 0x0000000002b708ed clang::EmitBackendOutput(clang::DiagnosticsEngine&, clang::CodeGenOptions const&, clang::TargetOptions const&, clang::LangOptions const&, llvm::StringRef, llvm::Module, clang::BackendAction, llvm::r\
aw_pwrite_stream) + 127
26 clang-3.7 0x0000000002b52282
27 clang-3.7 0x00000000035d13c0 clang::ParseAST(clang::Sema&, bool, bool) + 784
28 clang-3.7 0x00000000027e97c6 clang::ASTFrontendAction::ExecuteAction() + 322
29 clang-3.7 0x0000000002b54798 clang::CodeGenAction::ExecuteAction() + 1486
30 clang-3.7 0x00000000027e92a1 clang::FrontendAction::Execute() + 139
31 clang-3.7 0x00000000027b0ed4 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) + 772
32 clang-3.7 0x00000000028fe389 clang::ExecuteCompilerInvocation(clang::CompilerInstance) + 993
33 clang-3.7 0x00000000012d4a22 cc1_main(llvm::ArrayRef<char const>, char const, void) + 770
34 clang-3.7 0x00000000012cd28e
35 clang-3.7 0x00000000012cd875 main + 1074
36 libc.so.6 0x00002b5d51ab5ec5 __libc_start_main + 245
37 clang-3.7 0x00000000012ca239
Stack dump:
PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT: Preprocessed source(s) and associated run script(s) are located at: clang-3.7: note: diagnostic msg: /tmp/filterbanks_neon-c2acf1.c clang-3.7: note: diagnostic msg: /tmp/filterbanks_neon-c2acf1.sh clang-3.7: note: diagnostic msg:
make: *** [trunk-clang] Error 254