[LLVM] llvm-reduce的使用

llvm-reduce
用来简化 IR 级别的用例。
其根据 执行命令的成功与失败来对用例进行简化。
使用很简单：

写好执行命令

$cat run.sh
#!/bin/bash
llc $1 2>&1 | grep "build failed"

执行

$llvm-reduce --test=run.sh test.ll

 **** SUCCESS | lines: 0
Saved new best reduction to reduced.ll
Couldn't increase anymore.

Done reducing! Reduced testcase: reduced.ll

最后输出简化后用例到 reduced.ll

遇到的问题：
1 起初我把脚本中的 “llc $1” 改为 “llc test.ll”，但llvm-reduce 命令行还是不能省略test.ll。
而且即便加了test.ll 居然没有作用。
起初的失败demo：

$cat run.sh
#!/bin/bash
llc test.ll 2>&1 | grep "build failed"

$llvm-reduce --test=run.sh           // fail
$llvm-reduce --test=run.sh test.ll   // output fail

llvm-reduce options

llvm-reduce --help
OVERVIEW: LLVM automatic testcase reducer.

USAGE: llvm-reduce [options] <input llvm ll/bc file>

OPTIONS:

Color Options:

  --color                             - Use colors in output (default=autodetect)

Generic Options:

  --help                              - Display available options (--help-hidden for more)
  --help-list                         - Display list of available options (--help-list-hidden for more)
  --version                           - Display the version of this program

llvm-reduce options:

  --abort-on-invalid-reduction        - Abort if any reduction results in invalid IR
  --delta-passes=<string>             - Delta passes to run, separated by commas. By default, run all delta passes.
  --in-place                          - WARNING: This option will replace your input file with the reduced version!
  --ir-passes=<string>                - A textual description of the pass pipeline, same as what's passed to `opt -passes`.
  -j <uint>                           - Maximum number of threads to use to process chunks. Set to 1 to disable parallelism.
  --max-pass-iterations=<int>         - Maximum number of times to run the full set of delta passes (default=5)
  --mtriple=<string>                  - Set the target triple
  --preserve-debug-environment        - Don't disable features used for crash debugging (crash reports, llvm-symbolizer and core dumps)
  --print-delta-passes                - Print list of delta passes, passable to --delta-passes as a comma separated list
  --skip-delta-passes=<string>        - Delta passes to not run, separated by commas. By default, run all delta passes.
  --starting-granularity-level=<uint> - Number of times to divide chunks prior to first test
  --test=<string>                     - Name of the interesting-ness test to be run
  --test-arg=<string>                 - Arguments passed onto the interesting-ness test
  --verbose                           - Print extra debugging information
  --write-tmp-files-as-bitcode        - Always write temporary files as bitcode instead of textual IR
  -x=<value>                          - Input language ('ir' or 'mir')
    =ir
    =mir

run 个别 passes

比如我们只想run function-bodies,instructions 这2个passes，可以

llvm-reduce --delta-passes="function-bodies,instructions" --test=run.sh test.ll

llvm-reduce的pass 主要有

// llvm/tools/llvm-reduce/DeltaManager.cpp
#define DELTA_PASSES                                                           \
  do {                                                                         \
    DELTA_PASS("strip-debug-info", stripDebugInfoDeltaPass)                    \
    DELTA_PASS("functions", reduceFunctionsDeltaPass)                          \
    DELTA_PASS("function-bodies", reduceFunctionBodiesDeltaPass)               \
    DELTA_PASS("special-globals", reduceSpecialGlobalsDeltaPass)               \
    DELTA_PASS("aliases", reduceAliasesDeltaPass)                              \
    DELTA_PASS("ifuncs", reduceIFuncsDeltaPass)                                \
    DELTA_PASS("simplify-conditionals-true", reduceConditionalsTrueDeltaPass)  \
    DELTA_PASS("simplify-conditionals-false",                                  \
               reduceConditionalsFalseDeltaPass)                               \
    DELTA_PASS("invokes", reduceInvokesDeltaPass)                              \
    DELTA_PASS("unreachable-basic-blocks",                                     \
               reduceUnreachableBasicBlocksDeltaPass)                          \
    DELTA_PASS("basic-blocks", reduceBasicBlocksDeltaPass)                     \
    DELTA_PASS("simplify-cfg", reduceUsingSimplifyCFGDeltaPass)                \
    DELTA_PASS("function-data", reduceFunctionDataDeltaPass)                   \
    DELTA_PASS("global-values", reduceGlobalValuesDeltaPass)                   \
    DELTA_PASS("global-objects", reduceGlobalObjectsDeltaPass)                 \
    DELTA_PASS("global-initializers", reduceGlobalsInitializersDeltaPass)      \
    DELTA_PASS("global-variables", reduceGlobalsDeltaPass)                     \
    DELTA_PASS("di-metadata", reduceDIMetadataDeltaPass)                       \
    DELTA_PASS("dpvalues", reduceDPValuesDeltaPass)                            \
    DELTA_PASS("metadata", reduceMetadataDeltaPass)                            \
    DELTA_PASS("named-metadata", reduceNamedMetadataDeltaPass)                 \
    DELTA_PASS("arguments", reduceArgumentsDeltaPass)                          \
    DELTA_PASS("instructions", reduceInstructionsDeltaPass)                    \
    DELTA_PASS("simplify-instructions", simplifyInstructionsDeltaPass)         \
    DELTA_PASS("ir-passes", runIRPassesDeltaPass)                              \
    DELTA_PASS("operands-zero", reduceOperandsZeroDeltaPass)                   \
    DELTA_PASS("operands-one", reduceOperandsOneDeltaPass)                     \
    DELTA_PASS("operands-nan", reduceOperandsNaNDeltaPass)                     \
    DELTA_PASS("operands-to-args", reduceOperandsToArgsDeltaPass)              \
    DELTA_PASS("operands-skip", reduceOperandsSkipDeltaPass)                   \
    DELTA_PASS("operand-bundles", reduceOperandBundesDeltaPass)                \
    DELTA_PASS("attributes", reduceAttributesDeltaPass)                        \
    DELTA_PASS("module-data", reduceModuleDataDeltaPass)                       \
    DELTA_PASS("opcodes", reduceOpcodesDeltaPass)                              \
    DELTA_PASS("volatile", reduceVolatileInstructionsDeltaPass)                \
    DELTA_PASS("atomic-ordering", reduceAtomicOrderingDeltaPass)               \
    DELTA_PASS("syncscopes", reduceAtomicSyncScopesDeltaPass)                  \
    DELTA_PASS("instruction-flags", reduceInstructionFlagsDeltaPass)           \
} while (false)

#define DELTA_PASSES_MIR                                                       \
  do {                                                                         \
    DELTA_PASS("instructions", reduceInstructionsMIRDeltaPass)                 \
    DELTA_PASS("ir-instruction-references",                                    \
               reduceIRInstructionReferencesDeltaPass)                         \
    DELTA_PASS("ir-block-references", reduceIRBlockReferencesDeltaPass)        \
    DELTA_PASS("ir-function-references", reduceIRFunctionReferencesDeltaPass)  \
    DELTA_PASS("instruction-flags", reduceInstructionFlagsMIRDeltaPass)        \
    DELTA_PASS("register-uses", reduceRegisterUsesMIRDeltaPass)                \
    DELTA_PASS("register-defs", reduceRegisterDefsMIRDeltaPass)                \
    DELTA_PASS("register-hints", reduceVirtualRegisterHintsDeltaPass)          \
    DELTA_PASS("register-masks", reduceRegisterMasksMIRDeltaPass)              \
  } while (false)

定制化llvm-reduce

此外我们还可以修改llvm-reduce的code来定制化我们的reduce，比如不想在reduce时删除 带meta为range的指令：

diff --git a/llvm/tools/llvm-reduce/deltas/ReduceMetadata.cpp b/llvm/tools/llvm-reduce/deltas/ReduceMetadata.cpp
index 30bf612fe969..73fb2cd2fd00 100644
--- a/llvm/tools/llvm-reduce/deltas/ReduceMetadata.cpp
+++ b/llvm/tools/llvm-reduce/deltas/ReduceMetadata.cpp
@@ -21,11 +21,15 @@
 using namespace llvm;

 static bool shouldKeepDebugIntrinsicMetadata(Instruction &I, MDNode &MD) {
+  const MDNode *Range = I.getMetadata(LLVMContext::MD_range);
+  if (Range)
+    return true;
   return isa<DILocation>(MD) && isa<DbgInfoIntrinsic>(I);
 }