Jeremy Morse via llvm-dev
2020-Nov-11 11:13 UTC
[llvm-dev] [RFC] A value-tracking LiveDebugValues implementation
Hi Xiang, On Wed, Nov 11, 2020 at 1:59 AM Zhang, Xiang1 <xiang1.zhang at intel.com> wrote:> Jeremy wrote: > > ... The value %0 is live up to and including the ADD64ri but not past it, meaning LLVM today will drop the DBG_VALUE ... > > Just a little puzzle about the " drop the DBG_VALUE ", maybe I didn't get your key point, > So I write a phi case, it like > ___________________________________________________ > vaarg.end: ; preds = %vaarg.in_mem, %vaarg.in_reg > %5 = phi i32 [ %4, %vaarg.in_reg ], [ %l5, %vaarg.in_mem ], !dbg !50 > %6 = add nsw i32 %gp_offset, %vaarg.l, !dbg !41 > call void @llvm.dbg.value(metadata i32 %5, metadata !28, metadata !DIExpression()), !dbg !50 > ret i32 %6, !dbg !52 > ____________________________________________________ > After phi-elimination, it is > > bb.3.vaarg.end: > ; predecessors: %bb.2, %bb.1 > > %5:gr32 = COPY killed %18:gr32, debug-location !50; phi.ll:21:1 // %18 was def in all predecessors //phi --> COPY killed %18 > %17:gr32 = ADD32rr killed %8:gr32(tied-def 0), killed %5:gr32, implicit-def dead $eflags, debug-location !41; phi.ll:12:1 // kill %5 > DBG_VALUE %5:gr32, $noreg, !"15", !DIExpression(), debug-location !50; phi.ll:21:1 line no:21 > $eax = COPY killed %17:gr32, debug-location !51; phi.ll:23:1 > RETQ implicit killed $eax, debug-location !51; phi.ll:23:1 > ____________________________________________________ > Go on, After Simple Register Coalescing > > 336B bb.3.vaarg.end: > ; predecessors: %bb.2, %bb.1 > > 384B %17:gr32 = ADD32rr %17:gr32(tied-def 0), %18:gr32, implicit-def dead $eflags, debug-location !41; phi.ll:12:1 > DBG_VALUE %18:gr32, $noreg, !"15", !DIExpression(), debug-location !50; phi.ll:21:1 line no:21 // Seems still works here. (not "drop the DBG_VALUE ") > 400B $eax = COPY %17:gr32, debug-location !51; phi.ll:23:1 > 416B RETQ implicit killed $eax, debug-location !51; phi.ll:23:1 > ____________________________________________________Indeed; it's LiveDebugVariables that drops out-of-liveness DBG_VALUEs, as seen here: https://github.com/llvm/llvm-project/blob/68ac02c0dd2b8fda52ac132a86f72f2ad6b139a5/llvm/lib/CodeGen/LiveDebugVariables.cpp#L657 as the register allocator will not maintain the location of its vreg for positions that aren't live. -- Thanks, Jeremy
Zhang, Xiang1 via llvm-dev
2020-Nov-11 11:31 UTC
[llvm-dev] [RFC] A value-tracking LiveDebugValues implementation
Got it! Thank you very much!! : ) -----Original Message----- From: Jeremy Morse <jeremy.morse.llvm at gmail.com> Sent: Wednesday, November 11, 2020 7:13 PM To: Zhang, Xiang1 <xiang1.zhang at intel.com> Cc: Adrian Prantl <aprantl at apple.com>; Paul Robinson <paul.robinson at sony.com>; Eric Christopher <echristo at gmail.com>; Jonas Devlieghere <jdevlieghere at apple.com>; David Blaikie <dblaikie at gmail.com>; Vedant Kumar <vedant_kumar at apple.com>; Reid Kleckner <rnk at google.com>; orlando.hyams at sony.com; Stephen.Tozer at sony.com; Djordje Todorovic <djordje.todorovic at rt-rk.com>; Luo, Yuanke <yuanke.luo at intel.com>; Zhang, Annita <annita.zhang at intel.com>; Wang, Pengfei <pengfei.wang at intel.com>; llvm-dev <llvm-dev at lists.llvm.org> Subject: Re: [llvm-dev] [RFC] A value-tracking LiveDebugValues implementation Hi Xiang, On Wed, Nov 11, 2020 at 1:59 AM Zhang, Xiang1 <xiang1.zhang at intel.com> wrote:> Jeremy wrote: > > ... The value %0 is live up to and including the ADD64ri but not past it, meaning LLVM today will drop the DBG_VALUE ... > > Just a little puzzle about the " drop the DBG_VALUE ", maybe I didn't get your key point, > So I write a phi case, it like > ___________________________________________________ > vaarg.end: ; preds = %vaarg.in_mem, %vaarg.in_reg > %5 = phi i32 [ %4, %vaarg.in_reg ], [ %l5, %vaarg.in_mem ], !dbg !50 > %6 = add nsw i32 %gp_offset, %vaarg.l, !dbg !41 > call void @llvm.dbg.value(metadata i32 %5, metadata !28, metadata !DIExpression()), !dbg !50 > ret i32 %6, !dbg !52 > ____________________________________________________ > After phi-elimination, it is > > bb.3.vaarg.end: > ; predecessors: %bb.2, %bb.1 > > %5:gr32 = COPY killed %18:gr32, debug-location !50; phi.ll:21:1 // %18 was def in all predecessors //phi --> COPY killed %18 > %17:gr32 = ADD32rr killed %8:gr32(tied-def 0), killed %5:gr32, implicit-def dead $eflags, debug-location !41; phi.ll:12:1 // kill %5 > DBG_VALUE %5:gr32, $noreg, !"15", !DIExpression(), debug-location !50; phi.ll:21:1 line no:21 > $eax = COPY killed %17:gr32, debug-location !51; phi.ll:23:1 > RETQ implicit killed $eax, debug-location !51; phi.ll:23:1 > ____________________________________________________ > Go on, After Simple Register Coalescing > > 336B bb.3.vaarg.end: > ; predecessors: %bb.2, %bb.1 > > 384B %17:gr32 = ADD32rr %17:gr32(tied-def 0), %18:gr32, implicit-def dead $eflags, debug-location !41; phi.ll:12:1 > DBG_VALUE %18:gr32, $noreg, !"15", !DIExpression(), debug-location !50; phi.ll:21:1 line no:21 // Seems still works here. (not "drop the DBG_VALUE ") > 400B $eax = COPY %17:gr32, debug-location !51; phi.ll:23:1 > 416B RETQ implicit killed $eax, debug-location !51; phi.ll:23:1 > ____________________________________________________Indeed; it's LiveDebugVariables that drops out-of-liveness DBG_VALUEs, as seen here: https://github.com/llvm/llvm-project/blob/68ac02c0dd2b8fda52ac132a86f72f2ad6b139a5/llvm/lib/CodeGen/LiveDebugVariables.cpp#L657 as the register allocator will not maintain the location of its vreg for positions that aren't live. -- Thanks, Jeremy