给出函数内的以下C#代码:
....
var documentCollection =
client.CreateDocumentCollectionQuery("dbs/" + database.Id)
.Where(c => c.Id == DocumentCollectionName)
.AsEnumerable()
.FirstOrDefault();
if (documentCollection == null)
{
documentCollection =
await
client.CreateDocumentCollectionAsync(
"dbs/" + database.Id,
new DocumentCollection { Id = DocumentCollectionName });
}
return client;
注意:我没有返回documentCollection,我只需要初始化,如果还没有(CreateDocumentCollectionAsync调用).所以 - 在if块之后,documentCollection变成一个未使用的变量.
现在 - ReSharper建议将其优化为:
var documentCollection =
client.CreateDocumentCollectionQuery("dbs/" + database.Id)
.Where(c => c.Id == DocumentCollectionName)
.AsEnumerable()
.FirstOrDefault()
?? await
client.CreateDocumentCollectionAsync(
"dbs/" + database.Id,
new DocumentCollection { Id = DocumentCollectionName });
并且现在表明这documentCollection是一个未使用的变量.
我的问题: C#代码优化还是'发布'版本会完全删除这行代码并导致CreateDocumentCollectionAsync永不触发？
C#优化课程告诉我,'release'会在函数中不需要"下线"时构建垃圾收集变量,而调试版本则不会这样做(出于调试目的).
我现在想知道它是否如此渴望,即使它优化了一个未使用的变量赋值(它在后台触发一个操作).
不,编译器和JIT都不会优化您的方法调用.
有一个JIT编译器的列表.它确实优化了if (false) { ... }例如块或未使用的变量赋值.它不仅可以优化您的方法调用.如果这是真的,那么对void方法的每次调用都应该消失.
没有.
任何优化器都只能删除没有可观察行为的代码.
否则它不是一个优化者.
免责声明:这是实施细节,可能会有所变化,并带有一丝盐.
CLI规范的ECMA-335,第I.12.6.4节(优化)声明如下:
符合CLI的实现可以使用任何技术自由执行程序,这些技术在单个执行线程中保证线程生成的副作用和异常按照CIL指定的顺序可见.为此目的,仅挥发性操作(包括挥发性读取)构成可见的副作用.(请注意,虽然只有易失性操作构成可见的副作用,但易失性操作也会影响非易失性引用的可见性.)易失性操作在§I.12.6.7中指定.相对于另一个线程注入线程的异常,没有排序保证(此类异常有时称为"异步异常"(例如,System.Threading.ThreadAbortException).
[基本原理:优化编译器可以自由地重新排序副作用和同步异常,只要这种重新排序不会改变任何可观察的程序行为.最终理由]
[注意:允许CLI的实现使用优化编译器,例如,将CIL转换为本机机器代码,前提是编译器维护(在每个执行线程内)相同的副作用和同步异常顺序.这是比ISO C++(允许在一对序列点之间重新排序)或ISO Scheme(允许对函数的参数重新排序)更强的条件.结束说明]
这意味着符合CLI的任何实现都可以自由地进行这样的优化,如果它可以保证副作用的顺序不会受到损害.这意味着如果方法没有副作用并且JIT或语言编译器静态地分析对于给定事实,则它可以优化它,因为在使用或不使用该方法的情况下将不会重新排序所述副作用.
话虽如此,目前,C#编译器将优化掉未使用的变量,但不会优化方法调用.编译器没有对整个方法调用进行静态分析,因此无法"证明"该方法在代码中没有副作用.更重要的是,JIT优化并不具有攻击性,它可能只是内联方法调用,而不是优化它.
作为开源,您可以看到x86 JIT编译阶段并查看正在进行的一些优化(via compphases.h):
// Names of x86 JIT phases, in order. Assumes that the caller defines CompPhaseNameMacro // in a useful way before including this file, e.g., to define the phase enumeration and the // corresponding array of string names of those phases. This include file undefines CompPhaseNameMacro // after the last use. // The arguments are: // CompPhaseNameMacro(enumName, stringName, hasChildren, parent) // "enumName" is an Enumeration-style all-caps name. // "stringName" is a self-explanatory. // "hasChildren" is true if this phase is broken out into subphases. // (We should never do EndPhase on a phase that has children, only on 'leaf phases.') // "parent" is -1 for leaf phases, otherwise it is the "enumName" of the parent phase. CompPhaseNameMacro(PHASE_PRE_IMPORT, "Pre-import", "PRE-IMP", false, -1) CompPhaseNameMacro(PHASE_IMPORTATION, "Importation", "IMPORT", false, -1) CompPhaseNameMacro(PHASE_POST_IMPORT, "Post-import", "POST-IMP", false, -1) CompPhaseNameMacro(PHASE_MORPH, "Morph", "MORPH", false, -1) CompPhaseNameMacro(PHASE_GS_COOKIE, "GS Cookie", "GS-COOK", false, -1) CompPhaseNameMacro(PHASE_COMPUTE_PREDS, "Compute preds", "PREDS", false, -1) CompPhaseNameMacro(PHASE_MARK_GC_POLL_BLOCKS, "Mark GC poll blocks", "GC-POLL", false, -1) CompPhaseNameMacro(PHASE_COMPUTE_EDGE_WEIGHTS, "Compute edge weights (1)", "EDG-WGT", false, -1) #if FEATURE_EH_FUNCLETS CompPhaseNameMacro(PHASE_CREATE_FUNCLETS, "Create EH funclets", "EH-FUNC", false, -1) #endif // FEATURE_EH_FUNCLETS CompPhaseNameMacro(PHASE_OPTIMIZE_LAYOUT, "Optimize layout", "LAYOUT", false, -1) CompPhaseNameMacro(PHASE_OPTIMIZE_LOOPS, "Optimize loops", "LOOP-OPT", false, -1) CompPhaseNameMacro(PHASE_CLONE_LOOPS, "Clone loops", "LP-CLONE", false, -1) CompPhaseNameMacro(PHASE_UNROLL_LOOPS, "Unroll loops", "UNROLL", false, -1) CompPhaseNameMacro(PHASE_HOIST_LOOP_CODE, "Hoist loop code", "LP-HOIST", false, -1) CompPhaseNameMacro(PHASE_MARK_LOCAL_VARS, "Mark local vars", "MARK-LCL", false, -1) CompPhaseNameMacro(PHASE_OPTIMIZE_BOOLS, "Optimize bools", "OPT-BOOL", false, -1) CompPhaseNameMacro(PHASE_FIND_OPER_ORDER, "Find oper order", "OPER-ORD", false, -1) CompPhaseNameMacro(PHASE_SET_BLOCK_ORDER, "Set block order", "BLK-ORD", false, -1) CompPhaseNameMacro(PHASE_BUILD_SSA, "Build SSA representation", "SSA", true, -1) CompPhaseNameMacro(PHASE_BUILD_SSA_TOPOSORT, "SSA: topological sort", "SSA-SORT", false, PHASE_BUILD_SSA) CompPhaseNameMacro(PHASE_BUILD_SSA_DOMS, "SSA: Doms1", "SSA-DOMS", false, PHASE_BUILD_SSA) CompPhaseNameMacro(PHASE_BUILD_SSA_LIVENESS, "SSA: liveness", "SSA-LIVE", false, PHASE_BUILD_SSA) CompPhaseNameMacro(PHASE_BUILD_SSA_IDF, "SSA: IDF", "SSA-IDF", false, PHASE_BUILD_SSA) CompPhaseNameMacro(PHASE_BUILD_SSA_INSERT_PHIS, "SSA: insert phis", "SSA-PHI", false, PHASE_BUILD_SSA) CompPhaseNameMacro(PHASE_BUILD_SSA_RENAME, "SSA: rename", "SSA-REN", false, PHASE_BUILD_SSA) CompPhaseNameMacro(PHASE_EARLY_PROP, "Early Value Propagation", "ERL-PROP", false, -1) CompPhaseNameMacro(PHASE_VALUE_NUMBER, "Do value numbering", "VAL-NUM", false, -1) CompPhaseNameMacro(PHASE_OPTIMIZE_INDEX_CHECKS, "Optimize index checks", "OPT-CHK", false, -1) #if FEATURE_VALNUM_CSE CompPhaseNameMacro(PHASE_OPTIMIZE_VALNUM_CSES, "Optimize Valnum CSEs", "OPT-CSE", false, -1) #endif CompPhaseNameMacro(PHASE_VN_COPY_PROP, "VN based copy prop", "CP-PROP", false, -1) #if ASSERTION_PROP CompPhaseNameMacro(PHASE_ASSERTION_PROP_MAIN, "Assertion prop", "AST-PROP", false, -1) #endif CompPhaseNameMacro(PHASE_UPDATE_FLOW_GRAPH, "Update flow graph", "UPD-FG", false, -1) CompPhaseNameMacro(PHASE_COMPUTE_EDGE_WEIGHTS2, "Compute edge weights (2)", "EDG-WGT2", false, -1) CompPhaseNameMacro(PHASE_DETERMINE_FIRST_COLD_BLOCK, "Determine first cold block", "COLD-BLK", false, -1) CompPhaseNameMacro(PHASE_RATIONALIZE, "Rationalize IR", "RAT", false, -1) CompPhaseNameMacro(PHASE_SIMPLE_LOWERING, "Do 'simple' lowering", "SMP-LWR", false, -1) CompPhaseNameMacro(PHASE_LCLVARLIVENESS, "Local var liveness", "LIVENESS", true, -1) CompPhaseNameMacro(PHASE_LCLVARLIVENESS_INIT, "Local var liveness init", "LIV-INIT", false, PHASE_LCLVARLIVENESS) CompPhaseNameMacro(PHASE_LCLVARLIVENESS_PERBLOCK,"Per block local var liveness", "LIV-BLK", false, PHASE_LCLVARLIVENESS) CompPhaseNameMacro(PHASE_LCLVARLIVENESS_INTERBLOCK, "Global local var liveness", "LIV-GLBL", false, PHASE_LCLVARLIVENESS) CompPhaseNameMacro(PHASE_LVA_ADJUST_REF_COUNTS, "LVA adjust ref counts", "REF-CNT", false, -1) #ifdef LEGACY_BACKEND CompPhaseNameMacro(PHASE_RA_ASSIGN_VARS, "RA assign vars", "REGALLOC", false, -1) #endif // LEGACY_BACKEND CompPhaseNameMacro(PHASE_LOWERING_DECOMP, "Lowering decomposition", "LWR-DEC", false, -1) CompPhaseNameMacro(PHASE_LOWERING, "Lowering nodeinfo", "LWR-INFO", false, -1) #ifndef LEGACY_BACKEND CompPhaseNameMacro(PHASE_LINEAR_SCAN, "Linear scan register alloc", "LSRA", true, -1) CompPhaseNameMacro(PHASE_LINEAR_SCAN_BUILD, "LSRA build intervals", "LSRA-BLD", false, PHASE_LINEAR_SCAN) CompPhaseNameMacro(PHASE_LINEAR_SCAN_ALLOC, "LSRA allocate", "LSRA-ALL", false, PHASE_LINEAR_SCAN) CompPhaseNameMacro(PHASE_LINEAR_SCAN_RESOLVE, "LSRA resolve", "LSRA-RES", false, PHASE_LINEAR_SCAN) #endif // !LEGACY_BACKEND CompPhaseNameMacro(PHASE_GENERATE_CODE, "Generate code", "CODEGEN", false, -1) CompPhaseNameMacro(PHASE_EMIT_CODE, "Emit code", "EMIT", false, -1) CompPhaseNameMacro(PHASE_EMIT_GCEH, "Emit GC+EH tables", "EMT-GCEH", false, -1)
一些优化是:
死代码消除
线性扫描寄存器分配
循环展开
范围检查消除
本文接着描述了JIT所做的一些优化,@ EricLippert总体上讨论了优化问题
京公网安备 11010802040832号 | 京ICP备19059560号-6 