BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Melbourne
X-LIC-LOCATION:Australia/Melbourne
BEGIN:DAYLIGHT
TZOFFSETFROM:+1000
TZOFFSETTO:+1100
TZNAME:AEDT
DTSTART:19721003T020000
RRULE:FREQ=YEARLY;BYMONTH=4;BYDAY=1SU
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19721003T020000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240214T070250Z
LOCATION:Meeting Room C4.11\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231215T111500
DTEND;TZID=Australia/Melbourne:20231215T113000
UID:siggraphasia_SIGGRAPH Asia 2023_sess136_papers_405@linklings.com
SUMMARY:SLANG.D: Fast, Modular and Differentiable Shader Programming
DESCRIPTION:Technical Papers\n\nSai Praveen Bangaru (MIT CSAIL), Lifan Wu 
 (NVIDIA), Tzu-Mao Li (University of California San Diego), Jacob Munkberg 
 (NVIDIA), Gilbert Bernstein (University of Washington), Jonathan Ragan-Kel
 ley (MIT CSAIL), Aaron Lefohn (NVIDIA), Fredo Durand (MIT CSAIL), and Yong
  He (NVIDIA)\n\nWe introduce SLANG.D, a shading language that incorporates
  first-class automatic differentiation support derived from the Slang lang
 uage. The new shading language allows us to transform a Direct3D-based pat
 h tracer to be fully differentiable with minor modifications to existing c
 ode. SLANG.D enables a shared ecosystem between machine learning framework
 s and pre-existing graphics hardware API-based rendering systems, promotin
 g the interchange of components and ideas across these two domains.\n\nOur
  contributions include a differentiable type system designed to ensure typ
 e safety and semantic clarity in codebases that blend differentiable and n
 on-differentiable code, language primitives that automatically generate bo
 th forward and reverse gradient propagation methods, and a compiler archit
 ecture that generates efficient derivative propagation shader code for gra
 phics pipelines. Our compiler supports differentiating code that involves 
 arbitrary control-flow, dynamic dispatch, generics and higher-order differ
 entiation, while providing developers flexible control of checkpointing an
 d gradient aggregation strategies for best performance. Our system allows 
 us to differentiate an existing real-time path tracer, Falcor, with minima
 l change to its shader code.\nWe show that the compiler-generated derivati
 ve kernels perform as efficiently as handwritten ones. In several benchmar
 ks, the SLANG.D code achieves significant speedup when compared to prior a
 utomatic differentiation systems.\n\nRegistration Category: Full Access\n\
 nSession Chair: Bo Ren (TMCC, Nankai University)
URL:https://asia.siggraph.org/2023/full-program?id=papers_405&sess=sess136
END:VEVENT
END:VCALENDAR