adrien/zig-dimal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Replace Scalar and Vector to a single Quantity that use @Vector #1

Merged
adrien merged 12 commits from simd into main 2026-04-25 23:29:30 +00:00
Conversation 0 Commits 12 Files Changed 10
+1496 -1713
12 Commits
Author SHA1 Message Date
adrien 4a2d45b384 Fix benchmarks to work with new Scalar and Vector 2026-04-26 01:28:33 +02:00
adrien d32de3fe82 Converted Base test to new Scalar 2026-04-26 01:18:41 +02:00
adrien 58fd4d5797 Fix
Deploy MkDocs to Garage / build-and-deploy (push) Failing after 29s
Details
2026-04-26 01:11:51 +02:00
adrien ae6b069da1 Added simd ref to action
Deploy MkDocs to Garage / build-and-deploy (push) Failing after 27s
Details
2026-04-26 01:09:20 +02:00
adrien 4b01dfe412 Replaced Vector with Quantity
Deploy MkDocs to Garage / build-and-deploy (push) Failing after 25s
Details
2026-04-26 01:07:12 +02:00
adrien c350ffd1d6 Working vector initiate
Deploy MkDocs to Garage / build-and-deploy (push) Failing after 30s
Details
2026-04-26 00:40:10 +02:00
adrien a591736b19 Totally replaced Scalar with Quantity
Deploy MkDocs to Garage / build-and-deploy (push) Failing after 49s
Details
2026-04-26 00:29:00 +02:00
adrien 89e46f4cd4 .
Deploy MkDocs to Garage / build-and-deploy (push) Failing after 2m17s
Details
2026-04-25 23:29:51 +02:00
adrien db6aaf433d Added mkdocs 2026-04-25 23:29:06 +02:00
adrien 0257ee5d21 Added Quantity that will be both Scalar and Vector, using only @Vector 
I did some benchmark, and @Vector(1, type) is just as fast as type for
float and around 40% slower for int.

But the traide off is ok for me, because for @Vector(>1, type), we are
between 2 and 32x faster! And I dont think single value int will be use
for high performance computing, so I am ok with how simple it make
everything.

Because now I dont have to one struct that use @Vector and one that use
just type, I can just use one. Making it even easier to integrate GPU
later. I will even be able to have a single place in my code with a + -
* and / basically. A single function for all, so I only need to optimize
that one.
 2026-04-25 22:36:48 +02:00
adrien 7fc7193206 FLush 2026-04-25 18:57:38 +02:00
adrien 3a4783cfa1 Benchmark SIMD vs Native 2026-04-25 18:47:13 +02:00