优化：循环变换 #58
Assignees
Comments
Cydiater
added a commit
that referenced
this issue
Aug 6, 2022
* merge IR::UnaryOp::ID * merge inst b=constexpr; c=a+b; * merge inst b=constexpr; c=a-b; * merge inst: a+b*4 * fix merge_inst * merge inst: a+c1*c2 * feat: backend dce * fix: backend dce * feat: optimize a[*|/|%](1<<c) * feat: optimize a/c * feat: optimize a%c * feat: optimize a*((1<<c1)[+|-](1<<c2)) * optimize GVN * warn * fix * fix * format * fix * feat: support LoadArg and ArrayIndex in gvn, simplify load store * feat: support LoadAddr in gvn * feat: dse in BB * run gvn aftere simple_load_store * fix: gvn constexpr compute * feat: simplify add expr * format * fix: simplify add expr * feat: support more general simplify_load_store on DAG subgraph of CFG * feat: merge pure func * fix: unsafe merge pure call * fix: side effect analysis * feat: const prop for functions that called once * feat: remove unused MemObject * feat: add remove_unused_BB * expose remove_unused_BB * feat: remove trivial jumps, and reorder code before backend * feat: remove unused return value * fix: remove LocalVarDef * feat: support phi identity in gvn * fix: remove trivial BB * feat: tail_rec_to_loop, to debug * fix: remove trivial BB * fix: tail_call_to_loop * do $bench * feat: simplify branch fix: ir exec phi fix: remove trivial bb fix: tail rec to loop * test: no inline * fix format * feat: add CondProp * fix: precondition of gcm & PointerBase feat: gvn remove phi like a=phi(b,b,a,a) * fix * feat: remove unreturnable BB $bench * feat: remove unused func (this fix CE on 95_float backend, but there may be other bugs) * feat: typecheck (float,int,addr) fix: float as condition $bench * fix: translate phi to mov * fix: disable dag_ir instead of assert(0) when type check Int/Addr fail $bench * feat: schedule phi * fix: [#106] ssa destruction $bench * $bench * fix: [#106] ssa_destruction $bench * feat: better RemoveUnusedStore (testcase sl) fix: frontend gen ArrayIndex by correct order $bench * chore: add .clang-format $bench * chore: split dag_ir.cpp into files * chore: split dag_ir.cpp into files * feat: add interface for visit DAG_IR::LoopTreeNode directly fix: update clang-format to 13 $bench * fix: use clang-format-13 $bench * fix: add gvn between global_var_to_local and mem2reg $bench * format * fix: reorder passes into correct order $bench * feat: inference loop var feat: (GlobalInitProp) propegate value of global init bug: GlobalInitProp for Float * feat: unroll: for(i=c1; i ? c2; i ??= c3) $bench * fix: optimize_ir $bench * fix: simplify BB after loop unroll * feat: local array init to global feat: a[0] == a in gvn $bench * feat: global init to global $bench * feat: a%b = a-a/b*b $bench * refactor debug output * feat: loop unroll for(i=l;i<r;i+=1){...} => for(i=l;i+4<r;i+=4){...} for(;i<r;i+=4){...} feat: check IR properties everywhere $bench * test unroll(2) $bench * change exit hash to debug * fix CE * fix CE * feat: use args to set unroll parameters change exit hash to debug * disable optimizer for some functional testcases $bench * feat: inference const MemObject fix: SideEffect::mustbe * feat: const not in may_read $bench Co-authored-by: Cydiater <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
for(int i=0;i<n;++i)可以确定i的范围是0~n-1）...
The text was updated successfully, but these errors were encountered: