Bootstrap the new SIMD module #255
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
By using `_mm_set_epi16` we can set the value using a single instruction instead of three. With this change, we can get rid of the MMX instructions `_mm_setzero_si64` and `_mm_set_pi16`. Refs: scality#222
First, add some useful instruction set-specific definitions. Two types: - RegisterType: underlying representation of the future Register class. - MaskType: same as RegisterType, but for the Mask class. Three useful constants: - INSTRUCTION_SET: useful for discrimination. - ALIGNMENT: useful for memory allocation. - REG_BITSZ: useful for iteration. Refs: scality#222
This first test is pretty dumb, but it will be useful to bootstrap and setup the CircleCI pipeline. Refs: scality#222
lamphamsy
requested changes
Oct 16, 2018
| // We required AVX2 because we relies on some instructions (such as | ||
| // `_mm256_add_epi16` and others) that aren't available in the first version of | ||
| // AVX. | ||
| #if defined(__AVX__) && defined(__AVX2__) |
There was a problem hiding this comment.
Is defined(__AVX2__) enough?
There was a problem hiding this comment.
Yes, it's enough but I prefer to be explicit: here we want AVX BUT because of QuadIron uses of SIMD we required AVX2.
There was a problem hiding this comment.
So I think we could simplify it, also the comment as we need AVX2
There was a problem hiding this comment.
We could, but I think explicit is better than implicit (and removing AVX doesn't really make the code more or less readable)
src/simd_nf4.h
Outdated
| @@ -52,19 +54,18 @@ static inline aint128 m128i_to_uint128(m128i v) | |||
| inline aint128 expand16(uint16_t* arr, int n) | |||
| { | |||
| // since n <= 4 | |||
| uint16_t _arr[4] __attribute__((aligned(ALIGN_SIZE))) = {0, 0, 0, 0}; | |||
| uint16_t _arr[4] __attribute__((aligned(simd::ALIGNMENT))) = {0, 0, 0, 0}; | |||
There was a problem hiding this comment.
It is in simd namespace so simd:: is not necessary
There was a problem hiding this comment.
Good catch! I'll fix this 😃
That way, it can be ran in parallel with the test suite. Refs: scality#251
slaperche-scality
force-pushed
the
eh/bootstrap_new_simd_api
branch
from
7ce4d38 to
83e2bec
Compare
Now, we are sure that every codepath is covered and we won't silently break one of them. Refs: scality#222, scality#251
slaperche-scality
force-pushed
the
eh/bootstrap_new_simd_api
branch
3 times, most recently
from
d141979 to
35ad3a8
Compare
With this allocator we can use the STL container, such as std::vector, and be sure that we can use aligned load instruction when necessary. Refs: scality#222
The new implementation is more C++-like (can be used with STL container) and more robust (overflow handling, strict aliasing rule, …). Refs: scality#222
slaperche-scality
force-pushed
the
eh/bootstrap_new_simd_api
branch
from
35ad3a8 to
ba0ec36
Compare
lamphamsy
reviewed
Oct 17, 2018
lamphamsy
approved these changes
Oct 17, 2018
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.
This PR just lay the foundation for the new SIMD module, the biggest part will come in the next PR.
That being said, here we defines the basic constants and helpers, as well as a custom allocator.