Hi, Adding a new way that changes heap closure representation won't be trivial. Handling of the profiling way is very ad-hoc. Did you update the `deriveConstants` utils and `ghc:GHC.Runtime.Heap.Layout.*HdrSize` to take into account your new optional field? Sylvain On 18/03/2024 08:47, Adithya Kumar wrote:
Hey folks,
I'm adding a field to "StgHeader" under a flag.
``` typedef struct { const StgInfoTable* info; #if defined(NEW_WAY) StgWord new_field; #endif #if defined(PROFILING) StgProfHeader prof; #endif } StgHeader;
typedef struct { const StgInfoTable* info; #if defined(NEW_WAY) StgWord new_field; #endif #if defined(PROFILING) StgProfHeader prof; #endif StgSMPThunkHeader smp; } StgThunkHeader; ```
Follow-up changes: ==================
The following are the additional changes I've made following the `PROFILING` flag:
Similar to `PROF_HDR_FIELDS` macro, I've added `NEW_HDR_FIELDS` macro:
``` #if defined(NEW_WAY) #define NEW_HDR_FIELDS(w_,hdr) w_ hdr, #else #define NEW_HDR_FIELDS(w_,hdr) /* Nothing */ #endif ``` and updated the required stg code accordingly following the changes from `PROF_HDR_FIELDS`.
I've also updated `GHC.Exts.Heap.ClosureTypes.closureTypeHeaderSize` accordingly.
Error: ======
I'm greeted with the following error, ``` ASSERTION FAILED: file rts/StgStartup.cmm, line 100 ```
Basically, the following assert fails ``` #define CHECK_SENSIBLE_REGS() \ ASSERT(Hp != 0); \ ASSERT(HpAlloc == 0); \ ASSERT(Sp != 0); \ ASSERT(SpLim != 0); \ ASSERT(SpLim - WDS(RESERVED_STACK_WORDS) <= Sp); ```
More context for the error: ===========================
Values for the correct version: ``` Hp 283468906488 HpAlloc 0 Sp 283468911440 SpLim 283468910784 SpLim - WDS(RESERVED_STACK_WORDS) 1069080 ```
Values with the changes: ``` Hp 283468906488 HpAlloc 0 Sp 7532120 SpLim 9301536 SpLim - WDS(RESERVED_STACK_WORDS) 9301368 ```
I recorded these values by calling a C print from the Cmm code.
Clearly, my changes messed up a lot of things. How do I go about this now? The past 2 weeks I've only been looking at a lot of ghc code. If any more additional context is required I can provide it.
Best, Adithya
_______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs