Add PacketOps support for textures #329
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
rtabbara
force-pushed
the
packet_ops_tex_v2
branch
from
c5b76b9 to
fd01e20
Compare
wjakob
reviewed
Jan 27, 2025
rtabbara
added a commit
to mitsuba-renderer/mitsuba3
that referenced
this pull request
Jan 27, 2025
* Prior to Mitsuba 3.6, FP16 storage for bitmaps was disabled due to observed degradation in performance (particularly on LLVM backends) * Changes with Dr.Jit textures now using PacketOps (mitsuba-renderer/drjit#329) means we can re-enable support for FP16 bitmap textures
rtabbara
added a commit
to mitsuba-renderer/mitsuba3
that referenced
this pull request
Jan 27, 2025
* Prior to Mitsuba 3.6, FP16 storage for bitmaps was disabled due to observed degradation in performance (particularly on LLVM backends) * Changes with Dr.Jit textures now using PacketOps (mitsuba-renderer/drjit#329) means we can re-enable support for FP16 bitmap textures
rtabbara
force-pushed
the
packet_ops_tex_v2
branch
from
7883102 to
7d68dd6
Compare
njroussel
reviewed
Jan 29, 2025
rtabbara
force-pushed
the
packet_ops_tex_v2
branch
from
7d68dd6 to
41a01e4
Compare
Texture channel data is packed contiguously, so leverage PacketOps for gathers (and scatters) that show substantial speed-ups (particularly on the LLVM backend). This is especially true for half-precision textures, where the absence of FP16 SIMD gather instructions results in bloated compiled kernels and a degradation in performance relative to the single-precision counterpart. PacketOps circumvents this issue by avoiding the use of gather intrinsics entirely.
rtabbara
force-pushed
the
packet_ops_tex_v2
branch
from
41a01e4 to
6d58cfd
Compare
rtabbara
added a commit
to mitsuba-renderer/mitsuba3
that referenced
this pull request
Feb 5, 2025
* Prior to Mitsuba 3.6, FP16 storage for bitmaps was disabled due to observed degradation in performance (particularly on LLVM backends) * Changes with Dr.Jit textures now using PacketOps (mitsuba-renderer/drjit#329) means we can re-enable support for FP16 bitmap textures
njroussel
pushed a commit
to mitsuba-renderer/mitsuba3
that referenced
this pull request
Feb 6, 2025
* Prior to Mitsuba 3.6, FP16 storage for bitmaps was disabled due to observed degradation in performance (particularly on LLVM backends) * Changes with Dr.Jit textures now using PacketOps (mitsuba-renderer/drjit#329) means we can re-enable support for FP16 bitmap textures
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Texture channel data is packed contiguously, so leverage PacketOps for gathers (and scatters) that show substantial speed-ups (particularly on the LLVM backend).
This is especially true for half-precision textures, where the absence of FP16 SIMD gather instructions results in bloated compiled kernels and a degradation in performance relative to the single-precision counterpart. PacketOps circumvents this issue by avoiding the use of gather intrinsics entirely.