- Reenabled the optimization that load the symbols on demand only rather than loading all the symbols.
- Detected when a module loaded in the debug engine (dbghelp) misses the debug symbols, likely because it was implicitly loaded with an incomplete symbol search path and unloaded the module to reload it with the proper symbol search path set so that debug symbols could be found.
- For Editor, 'debug ensure' is significantly faste. The time to stack walk and dump went from ~20s to ~1s.
#jira UE-117236 - Ensure can freeze the Editor for 30+ seconds
#rb Johan.Berg
[FYI] Francis.Hurteau
#ROBOMERGE-SOURCE: CL 16603545 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v828-16531559)
[CL 16603547 by patrick laflamme in ue5-release-engine-test branch]
2. Add a project setting option to avoid generating velocity and TAA shader unexpectly for mobile platform.
#jira none
#rb Dmitriy.Dyomin, Guillaume.Abadie
#ROBOMERGE-SOURCE: CL 16584354 in //UE5/Main/...
#ROBOMERGE-BOT: STARSHIP (Main -> Release-Engine-Test) (v828-16531559)
[CL 16584358 by wei liu in ue5-release-engine-test branch]
2. Add a project setting option to avoid generating velocity and TAA shader unexpectly for mobile platform.
#jira none
#rb Dmitriy.Dyomin, Guillaume.Abadie
[CL 16584354 by Wei Liu in ue5-main branch]
#rb dan.phillips
#ROBOMERGE-OWNER: robert.millar
#ROBOMERGE-AUTHOR: robert.millar
#ROBOMERGE-SOURCE: CL 16554441 via CL 16554457 via CL 16554465 via CL 16554507 via CL 16554868
#ROBOMERGE-BOT: STARSHIP (Release-Engine-Test -> Main) (v828-16531559)
[CL 16554877 by robert millar in ue5-main branch]
[FYI] ben.woodhouse
#ROBOMERGE-SOURCE: CL 16554462 via CL 16554473 via CL 16554480 via CL 16554514 via CL 16554515
#ROBOMERGE-BOT: STARSHIP (Release-Engine-Test -> Main) (v828-16531559)
[CL 16554522 by robert millar in ue5-main branch]
Instruction latencies and uop counts from documentation manuals and community testing (referenced in the source), plus comparing assembly output on godbolt led to many shuffle and swizzle template overrides and specializations.
Also optimized access to the XY and ZW portions of the AVX m256 union to sub-optimal MSVC compiler output that turns them in to memory ops rather than simple casts or 1-latency extracts.
Added more shuffle/swizzle validation tests to UnrealMathTests to stress specific in-lane and cross-lane permutes.
#jira UE-107989
#codereview Dmytro.Vovk
[CL 16554362 by Zak Middleton in ue5-main branch]
