Bug 189535 - [WHLSL] The standard library should be fully tested
Summary: [WHLSL] The standard library should be fully tested
Status: RESOLVED MOVED
Alias: None
Product: WebKit
Classification: Unclassified
Component: New Bugs (show other bugs)
Version: WebKit Nightly Build
Hardware: Unspecified Unspecified
: P2 Normal
Assignee: Thomas Denney
URL:
Keywords:
Depends on:
Blocks: 176199
  Show dependency tree / graph
 
Reported: 2018-09-11 18:54 PDT by Thomas Denney
Modified: 2018-10-13 14:54 PDT (History)
2 users (show)

See Also:

Attachments
WIP (1.37 KB, patch)
2018-09-11 18:54 PDT, Thomas Denney 		no flags Details | Formatted Diff | Diff
View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Thomas Denney 2018-09-11 18:54:09 PDT 
[WHLSL] The standard library should be fully tested
Comment 1 Thomas Denney 2018-09-11 18:54:46 PDT 
Created attachment 349500 [details]
WIP
Comment 2 Thomas Denney 2018-09-11 19:00:05 PDT 
We cannot remove the preparation stage that type checks the standard library in full (which makes running tests *much* faster) without ensuring that every function in the standard library is tested to ensure that each standard library function actually gets type checked.
Comment 3 Myles C. Maxfield 2018-09-21 18:16:23 PDT 
There are a couple functions that I'm not very confident that their behavior matches HLSL:

firstbithigh()
firstbitlow()
countbits()
reversebits()
all the texturing functions

Also, these functions are implemented in a suboptimal way:

log2()
log10()
exp2()
fmod()
isordered()

Also: fma() and mad() are implemented identically, and I'm not sure if this is right.
Comment 4 Thomas Denney 2018-09-21 18:57:18 PDT 
(In reply to Myles C. Maxfield from comment #3)
> There are a couple functions that I'm not very confident that their behavior
> matches HLSL:
> 
> firstbithigh()
> firstbitlow()
> countbits()
> reversebits()
> all the texturing functions
> 
> Also, these functions are implemented in a suboptimal way:
> 
> log2()
> log10()
> exp2()
> fmod()
> isordered()
> 
> Also: fma() and mad() are implemented identically, and I'm not sure if this
> is right.

Likewise I’m pretty sure that isnormal isn’t implement correctly (it currently returns isNaN).
Comment 5 Myles C. Maxfield 2018-10-13 14:30:56 PDT 
Migrated to https://api.github.com/repos/gpuweb/WHLSL/issues/36
Comment 6 Myles C. Maxfield 2018-10-13 14:54:37 PDT 
https://github.com/gpuweb/WHLSL/issues/36