Closed
Bug 1910423
Opened 1 year ago
Closed 1 year ago
Implement Float64 -> Float16 conversion with inline assembly
Categories
(Core :: JavaScript Engine: JIT, enhancement, P1)
Core
JavaScript Engine: JIT
Tracking
()
RESOLVED
FIXED
131 Branch
Tracking | Status | |
---|---|---|
firefox131 | --- | fixed |
People
(Reporter: anba, Assigned: anba)
References
(Blocks 1 open bug)
Details
Attachments
(2 files)
Implement Float64 -> Float16 conversion without a call using the technique linked from https://github.com/tc39/proposal-float16array/issues/12#issuecomment-2026271444.
Assignee | ||
Comment 1•1 year ago
|
||
Used in the next part because it gives a noticeable performance boost
when we can avoid extra GPR<>XMM register moves.
Assignee | ||
Comment 2•1 year ago
|
||
Add an inline assembly path to convert Float64 directly to Float16 without a call
into the runtime. Doubles the performances when converting a double to float16.
Updated•1 year ago
|
Pushed by andre.bargull@gmail.com:
https://hg.mozilla.org/integration/autoland/rev/e3b75650396f
Part 1: Support encoding vpsignd instruction. r=jandem
https://hg.mozilla.org/integration/autoland/rev/988852519e9f
Part 2: Add inline Float64 -> Float16 conversion through Float32. r=jandem
Comment 4•1 year ago
|
||
bugherder |
https://hg.mozilla.org/mozilla-central/rev/e3b75650396f
https://hg.mozilla.org/mozilla-central/rev/988852519e9f
Status: ASSIGNED → RESOLVED
Closed: 1 year ago
status-firefox131:
--- → fixed
Resolution: --- → FIXED
Target Milestone: --- → 131 Branch
You need to log in
before you can comment on or make changes to this bug.
Description
•