[AArch64][SME2] Extend getRegAllocationHints for ZPRStridedOrContiguousReg #2

kmclaughlin-arm · 2024-11-15T17:33:00Z

ZPR2StridedOrContiguous loads used by a FORM_STRIDED_TUPLE pseudo
should attempt to assign a strided register to avoid unnecessary copies,
even though this may overlap with the list of SVE callee-saved registers.

…usReg ZPR2StridedOrContiguous loads used by a FORM_STRIDED_TUPLE pseudo should attempt to assign a strided register to avoid unnecessary copies, even though this may overlap with the list of SVE callee-saved registers.

…#116656) The main issue to solve is that OpenMP modifiers can be specified in any order, so the parser cannot expect any specific modifier at a given position. To solve that, define modifier to be a union of all allowable specific modifiers for a given clause. Additionally, implement modifier descriptors: for each modifier the corresponding descriptor contains a set of properties of the modifier that allow a common set of semantic checks. Start with the syntactic properties defined in the spec: Required, Unique, Exclusive, Ultimate, and implement common checks to verify each of them. OpenMP modifier overhaul: #2/3

…plementation (llvm#108413. llvm#117704) (llvm#117894) Relands llvm#117704, which relanded changes from llvm#108413 - this was reverted due to build issues. The new offload library did not build with `LIBOMPTARGET_OMPT_SUPPORT` enabled, which was not picked up by pre-merge testing. The last commit contains the fix; everything else is otherwise identical to the approved PR. ___ ### New API Previous discussions at the LLVM/Offload meeting have brought up the need for a new API for exposing the functionality of the plugins. This change introduces a very small subset of a new API, which is primarily for testing the offload tooling and demonstrating how a new API can fit into the existing code base without being too disruptive. Exact designs for these entry points and future additions can be worked out over time. The new API does however introduce the bare minimum functionality to implement device discovery for Unified Runtime and SYCL. This means that the `urinfo` and `sycl-ls` tools can be used on top of Offload. A (rough) implementation of a Unified Runtime adapter (aka plugin) for Offload is available [here](https://github.com/callumfare/unified-runtime/tree/offload_adapter). Our intention is to maintain this and use it to implement and test Offload API changes with SYCL. ### Demoing the new API ```sh # From the runtime build directory $ ninja LibomptUnitTests $ OFFLOAD_TRACE=1 ./offload/unittests/OffloadAPI/offload.unittests ``` ### Open questions and future work * Only some of the available device info is exposed, and not all the possible device queries needed for SYCL are implemented by the plugins. A sensible next step would be to refactor and extend the existing device info queries in the plugins. The existing info queries are all strings, but the new API introduces the ability to return any arbitrary type. * It may be sensible at some point for the plugins to implement the new API directly, and the higher level code on top of it could be made generic, but this is more of a long-term possibility.

…abort (llvm#117603) Hey guys, I found that Flang's built-in ABORT function is incomplete when I was using it. Compared with gfortran's ABORT (which can both abort and print out a backtrace), flang's ABORT implementation lacks the function of printing out a backtrace. This feature is essential for debugging and understanding the call stack at the failure point. To solve this problem, I completed the "// TODO:" of the abort function, and then implemented an additional built-in function BACKTRACE for flang. After a brief reading of the relevant source code, I used backtrace and backtrace_symbols in "execinfo.h" to quickly implement this. But since I used the above two functions directly, my implementation is slightly different from gfortran's implementation (in the output, the function call stack before main is additionally output, and the function line number is missing). In addition, since I used the above two functions, I did not need to add -g to embed debug information into the ELF file, but needed -rdynamic to ensure that the symbols are added to the dynamic symbol table (so that the function name will be printed out). Here is a comparison of the output between gfortran 's backtrace and my implementation: gfortran's implemention output: ``` #0 0x557eb71f4184 in testfun2_ at /home/hunter/plct/fortran/test.f90:5 #1 0x557eb71f4165 in testfun1_ at /home/hunter/plct/fortran/test.f90:13 #2 0x557eb71f4192 in test_backtrace at /home/hunter/plct/fortran/test.f90:17 llvm#3 0x557eb71f41ce in main at /home/hunter/plct/fortran/test.f90:18 ``` my impelmention output: ``` Backtrace: #0 ./test(_FortranABacktrace+0x32) [0x574f07efcf92] #1 ./test(testfun2_+0x14) [0x574f07efc7b4] #2 ./test(testfun1_+0xd) [0x574f07efc7cd] llvm#3 ./test(_QQmain+0x9) [0x574f07efc7e9] llvm#4 ./test(main+0x12) [0x574f07efc802] llvm#5 /usr/lib/libc.so.6(+0x25e08) [0x76954694fe08] llvm#6 /usr/lib/libc.so.6(__libc_start_main+0x8c) [0x76954694fecc] llvm#7 ./test(_start+0x25) [0x574f07efc6c5] ``` test program is: ``` function testfun2() result(err) implicit none integer :: err err = 1 call backtrace end function testfun2 subroutine testfun1() implicit none integer :: err integer :: testfun2 err = testfun2() end subroutine testfun1 program test_backtrace call testfun1() end program test_backtrace ``` I am well aware of the importance of line numbers, so I am now working on implementing line numbers (by parsing DWARF information) and supporting cross-platform (Windows) support.

…w API implementation (llvm#108413. llvm#117704)" (llvm#117995) Reverts llvm#117894 Buildbot failures in OpenMP/Offload bots. https://lab.llvm.org/buildbot/#/builders/30/builds/11193

…ne symbol size as symbols are created (llvm#117079)" This reverts commit ba668eb. Below test started failing again on x86_64 macOS CI. We're unsure if this patch is the exact cause, but since this patch has broken this test before, we speculatively revert it to see if it was indeed the root cause. ``` FAIL: lldb-shell :: Unwind/trap_frame_sym_ctx.test (1692 of 2162) ******************** TEST 'lldb-shell :: Unwind/trap_frame_sym_ctx.test' FAILED ******************** Exit Code: 1 Command Output (stderr): -- RUN: at line 7: /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/bin/clang --target=specify-a-target-or-use-a-_host-substitution --target=x86_64-apple-darwin22.6.0 -isysroot /Applications/Xcode-beta.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk -fmodules-cache-path=/Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/lldb-test-build.noindex/module-cache-clang/lldb-shell /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/Inputs/call-asm.c /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/Inputs/trap_frame_sym_ctx.s -o /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/tools/lldb/test/Shell/Unwind/Output/trap_frame_sym_ctx.test.tmp + /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/bin/clang --target=specify-a-target-or-use-a-_host-substitution --target=x86_64-apple-darwin22.6.0 -isysroot /Applications/Xcode-beta.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX.sdk -fmodules-cache-path=/Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/lldb-test-build.noindex/module-cache-clang/lldb-shell /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/Inputs/call-asm.c /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/Inputs/trap_frame_sym_ctx.s -o /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/tools/lldb/test/Shell/Unwind/Output/trap_frame_sym_ctx.test.tmp clang: warning: argument unused during compilation: '-fmodules-cache-path=/Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/lldb-test-build.noindex/module-cache-clang/lldb-shell' [-Wunused-command-line-argument] RUN: at line 8: /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/bin/lldb --no-lldbinit -S /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/tools/lldb/test/Shell/lit-lldb-init-quiet /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/tools/lldb/test/Shell/Unwind/Output/trap_frame_sym_ctx.test.tmp -s /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/trap_frame_sym_ctx.test -o exit | /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/bin/FileCheck /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/trap_frame_sym_ctx.test + /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/bin/lldb --no-lldbinit -S /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/tools/lldb/test/Shell/lit-lldb-init-quiet /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/tools/lldb/test/Shell/Unwind/Output/trap_frame_sym_ctx.test.tmp -s /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/trap_frame_sym_ctx.test -o exit + /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/lldb-build/bin/FileCheck /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/trap_frame_sym_ctx.test /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/trap_frame_sym_ctx.test:21:10: error: CHECK: expected string not found in input ^ <stdin>:26:64: note: scanning from here frame #1: 0x0000000100003ee9 trap_frame_sym_ctx.test.tmp`tramp ^ <stdin>:27:2: note: possible intended match here frame #2: 0x00007ff7bfeff6c0 ^ Input file: <stdin> Check file: /Users/ec2-user/jenkins/workspace/llvm.org/lldb-cmake/llvm-project/lldb/test/Shell/Unwind/trap_frame_sym_ctx.test -dump-input=help explains the following input dump. Input was: <<<<<< . . . 21: 0x100003ed1 <+0>: pushq %rbp 22: 0x100003ed2 <+1>: movq %rsp, %rbp 23: (lldb) thread backtrace -u 24: * thread #1, queue = 'com.apple.main-thread', stop reason = breakpoint 1.1 25: * frame #0: 0x0000000100003ecc trap_frame_sym_ctx.test.tmp`bar 26: frame #1: 0x0000000100003ee9 trap_frame_sym_ctx.test.tmp`tramp check:21'0 X error: no match found 27: frame #2: 0x00007ff7bfeff6c0 check:21'0 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ check:21'1 ? possible intended match 28: frame llvm#3: 0x0000000100003ec6 trap_frame_sym_ctx.test.tmp`main + 22 check:21'0 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 29: frame llvm#4: 0x0000000100003ec6 trap_frame_sym_ctx.test.tmp`main + 22 check:21'0 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 30: frame llvm#5: 0x00007ff8193cc41f dyld`start + 1903 check:21'0 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 31: (lldb) exit check:21'0 ~~~~~~~~~~~~ >>>>>> ```

## Description This PR fixes a segmentation fault that occurs when passing options requiring arguments via `-Xopenmp-target=<triple>`. The issue was that the function `Driver::getOffloadArchs` did not properly parse the extracted option, but instead assumed it was valid, leading to a crash when incomplete arguments were provided. ## Backtrace ```sh llvm-project/build/bin/clang++ main.cpp -fopenmp=libomp -fopenmp-targets=powerpc64le-ibm-linux-gnu -Xopenmp-target=powerpc64le-ibm-linux-gnu -o PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script. Stack dump: 0. Program arguments: llvm-project/build/bin/clang++ main.cpp -fopenmp=libomp -fopenmp-targets=powerpc64le-ibm-linux-gnu -Xopenmp-target=powerpc64le-ibm-linux-gnu -o 1. Compilation construction 2. Building compilation actions #0 0x0000562fb21c363b llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) (llvm-project/build/bin/clang+++0x392f63b) #1 0x0000562fb21c0e3c SignalHandler(int) Signals.cpp:0:0 #2 0x00007fcbf6c81420 __restore_rt (/lib/x86_64-linux-gnu/libpthread.so.0+0x14420) llvm#3 0x0000562fb1fa5d70 llvm::opt::Option::matches(llvm::opt::OptSpecifier) const (llvm-project/build/bin/clang+++0x3711d70) llvm#4 0x0000562fb2a78e7d clang::driver::Driver::getOffloadArchs(clang::driver::Compilation&, llvm::opt::DerivedArgList const&, clang::driver::Action::OffloadKind, clang::driver::ToolChain const*, bool) const (llvm-project/build/bin/clang+++0x41e4e7d) llvm#5 0x0000562fb2a7a9aa clang::driver::Driver::BuildOffloadingActions(clang::driver::Compilation&, llvm::opt::DerivedArgList&, std::pair<clang::driver::types::ID, llvm::opt::Arg const*> const&, clang::driver::Action*) const (.part.1164) Driver.cpp:0:0 llvm#6 0x0000562fb2a7c093 clang::driver::Driver::BuildActions(clang::driver::Compilation&, llvm::opt::DerivedArgList&, llvm::SmallVector<std::pair<clang::driver::types::ID, llvm::opt::Arg const*>, 16u> const&, llvm::SmallVector<clang::driver::Action*, 3u>&) const (llvm-project/build/bin/clang+++0x41e8093) llvm#7 0x0000562fb2a8395d clang::driver::Driver::BuildCompilation(llvm::ArrayRef<char const*>) (llvm-project/build/bin/clang+++0x41ef95d) llvm#8 0x0000562faf92684c clang_main(int, char**, llvm::ToolContext const&) (llvm-project/build/bin/clang+++0x109284c) llvm#9 0x0000562faf826cc6 main (llvm-project/build/bin/clang+++0xf92cc6) llvm#10 0x00007fcbf6699083 __libc_start_main /build/glibc-LcI20x/glibc-2.31/csu/../csu/libc-start.c:342:3 llvm#11 0x0000562faf923a5e _start (llvm-project/build/bin/clang+++0x108fa5e) [1] 2628042 segmentation fault (core dumped) main.cpp -fopenmp=libomp -fopenmp-targets=powerpc64le-ibm-linux-gnu -o ```

llvm#118923) …d reentry. These utilities provide new, more generic and easier to use support for lazy compilation in ORC. LazyReexportsManager is an alternative to LazyCallThroughManager. It takes requests for lazy re-entry points in the form of an alias map: lazy-reexports = { ( <entry point symbol #1>, <implementation symbol #1> ), ( <entry point symbol #2>, <implementation symbol #2> ), ... ( <entry point symbol #n>, <implementation symbol #n> ) } LazyReexportsManager then: 1. binds the entry points to the implementation names in an internal table. 2. creates a JIT re-entry trampoline for each entry point. 3. creates a redirectable symbol for each of the entry point name and binds redirectable symbol to the corresponding reentry trampoline. When an entry point symbol is first called at runtime (which may be on any thread of the JIT'd program) it will re-enter the JIT via the trampoline and trigger a lookup for the implementation symbol stored in LazyReexportsManager's internal table. When the lookup completes the entry point symbol will be updated (via the RedirectableSymbolManager) to point at the implementation symbol, and execution will proceed to the implementation symbol. Actual construction of the re-entry trampolines and redirectable symbols is delegated to an EmitTrampolines functor and the RedirectableSymbolsManager respectively. JITLinkReentryTrampolines.h provides a JITLink-based implementation of the EmitTrampolines functor. (AArch64 only in this patch, but other architectures will be added in the near future). Register state save and reentry functionality is added to the ORC runtime in the __orc_rt_sysv_resolve and __orc_rt_resolve_implementation functions (the latter is generic, the former will need custom implementations for each ABI and architecture to be supported, however this should be much less effort than the existing OrcABISupport approach, since the ORC runtime allows this code to be written as native assembly). The resulting system: 1. Works equally well for in-process and out-of-process JIT'd code. 2. Requires less boilerplate to set up. Given an ObjectLinkingLayer and PlatformJD (JITDylib containing the ORC runtime), setup is just: ```c++ auto RSMgr = JITLinkRedirectableSymbolManager::Create(OLL); if (!RSMgr) return RSMgr.takeError(); auto LRMgr = createJITLinkLazyReexportsManager(OLL, **RSMgr, PlatformJD); if (!LRMgr) return LRMgr.takeError(); ``` after which lazy reexports can be introduced with: ```c++ JD.define(lazyReexports(LRMgr, <alias map>)); ``` LazyObectLinkingLayer is updated to use this new method, but the LLVM-IR level CompileOnDemandLayer will continue to use LazyCallThroughManager and OrcABISupport until the new system supports a wider range of architectures and ABIs. The llvm-jitlink utility's -lazy option now uses the new scheme. Since it depends on the ORC runtime, the lazy-link.ll testcase and associated helpers are moved to the ORC runtime.

The Clang binary (and any binary linking Clang as a library), when built using PIE, ends up with a pretty shocking number of dynamic relocations to apply to the executable image: roughly 400k. Each of these takes up binary space in the executable, and perhaps most interestingly takes start-up time to apply the relocations. The largest pattern I identified were the strings used to describe target builtins. The addresses of these string literals were stored into huge arrays, each one requiring a dynamic relocation. The way to avoid this is to design the target builtins to use a single large table of strings and offsets within the table for the individual strings. This switches the builtin management to such a scheme. This saves over 100k dynamic relocations by my measurement, an over 25% reduction. Just looking at byte size improvements, using the `bloaty` tool to compare a newly built `clang` binary to an old one: ``` FILE SIZE VM SIZE -------------- -------------- +1.4% +653Ki +1.4% +653Ki .rodata +0.0% +960 +0.0% +960 .text +0.0% +197 +0.0% +197 .dynstr +0.0% +184 +0.0% +184 .eh_frame +0.0% +96 +0.0% +96 .dynsym +0.0% +40 +0.0% +40 .eh_frame_hdr +114% +32 [ = ] 0 [Unmapped] +0.0% +20 +0.0% +20 .gnu.hash +0.0% +8 +0.0% +8 .gnu.version +0.9% +7 +0.9% +7 [LOAD #2 [R]] [ = ] 0 -75.4% -3.00Ki .relro_padding -16.1% -802Ki -16.1% -802Ki .data.rel.ro -27.3% -2.52Mi -27.3% -2.52Mi .rela.dyn -1.6% -2.66Mi -1.6% -2.66Mi TOTAL ``` We get a 16% reduction in the `.data.rel.ro` section, and nearly 30% reduction in `.rela.dyn` where those reloctaions are stored. This is also visible in my benchmarking of binary start-up overhead at least: ``` Benchmark 1: ./old_clang --version Time (mean ± σ): 17.6 ms ± 1.5 ms [User: 4.1 ms, System: 13.3 ms] Range (min … max): 14.2 ms … 22.8 ms 162 runs Benchmark 2: ./new_clang --version Time (mean ± σ): 15.5 ms ± 1.4 ms [User: 3.6 ms, System: 11.8 ms] Range (min … max): 12.4 ms … 20.3 ms 216 runs Summary './new_clang --version' ran 1.13 ± 0.14 times faster than './old_clang --version' ``` We get about 2ms faster `--version` runs. While there is a lot of noise in binary execution time, this delta is pretty consistent, and represents over 10% improvement. This is particularly interesting to me because for very short source files, repeatedly starting the `clang` binary is actually the dominant cost. For example, `configure` scripts running against the `clang` compiler are slow in large part because of binary start up time, not the time to process the actual inputs to the compiler. ---- This PR implements the string tables using `constexpr` code and the existing macro system. I understand that the builtins are moving towards a TableGen model, and if complete that would provide more options for modeling this. Unfortunately, that migration isn't complete, and even the parts that are migrated still rely on the ability to break out of the TableGen model and directly expand an X-macro style `BUILTIN(...)` textually. I looked at trying to complete the move to TableGen, but it would both require the difficult migration of the remaining targets, and solving some tricky problems with how to move away from any macro-based expansion. I was also able to find a reasonably clean and effective way of doing this with the existing macros and some `constexpr` code that I think is clean enough to be a pretty good intermediate state, and maybe give a good target for the eventual TableGen solution. I was also able to factor the macros into set of consistent patterns that avoids a significant regression in overall boilerplate.

kmclaughlin-arm added 2 commits November 15, 2024 16:44

Add SME2 dot tests using the SVE calling convention

6cfe687

kmclaughlin-arm closed this Dec 13, 2024

kmclaughlin-arm deleted the sme2-svecc-regalloc-hints branch December 13, 2024 11:46

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[AArch64][SME2] Extend getRegAllocationHints for ZPRStridedOrContiguousReg #2

[AArch64][SME2] Extend getRegAllocationHints for ZPRStridedOrContiguousReg #2

kmclaughlin-arm commented Nov 15, 2024

[AArch64][SME2] Extend getRegAllocationHints for ZPRStridedOrContiguousReg #2

[AArch64][SME2] Extend getRegAllocationHints for ZPRStridedOrContiguousReg #2

Conversation

kmclaughlin-arm commented Nov 15, 2024