feat: add blas/base/strsm
#7021
Conversation
stdlib-bot
added
the
BLAS
ShabiShett07
added
GSoC
Coverage Report
The above coverage report was generated for the changes in this PR. |
| var f32 = require( '@stdlib/number/float64/base/to-float32' ); | ||
|
|
||
|
|
||
| // FUNCTIONS |
There was a problem hiding this comment.
| // FUNCTIONS | |
| // FUNCTIONS // |
| */ | ||
| function strsm( side, uplo, transa, diag, M, N, alpha, A, strideA1, strideA2, offsetA, B, strideB1, strideB2, offsetB ) { // eslint-disable-line max-params | ||
| var nonunit; | ||
| var isrma; |
There was a problem hiding this comment.
@ShabiShett07 Mind cleaning up these lint errors concerning indentation? By now you should have set up your local dev environment such that these sorts of lint errors never happen.
|
|
||
| // EXPORTS // | ||
|
|
||
| module.exports = strsm; |
| if ( isrmb ) { | ||
| // For row-major matrices, the last dimension has the fastest changing index... | ||
| sb0 = strideB2; // stride for innermost loop | ||
| sb1 = strideB1; // stride for outermost loop | ||
| } else { // isColMajor | ||
| // For column-major matrices, the first dimension has the fastest changing index... | ||
| sb0 = strideB1; // stride for innermost loop | ||
| sb1 = strideB2; // stride for outermost loop | ||
| } |
There was a problem hiding this comment.
This is why you are getting incorrect results. You need to do whatever A does above. You shouldn't independently swap the strides of B. When A and B are in two different orders, it is always the case that either A or B will be iterated in a non-cache optimal way. In which case, we just have to pick which matrix we want to optimize for and picking either A or B is valid. In this case, picking A is as good a choice as any.
Accordingly, L151-159 should become:
if ( isrma ) {
// For row-major matrices, the last dimension has the fastest changing index...
sa0 = strideA2; // stride for innermost loop
sa1 = strideA1; // stride for outermost loop
sb0 = strideB2;
sb1 = strideB1;
} else { // isColMajor
// For column-major matrices, the first dimension has the fastest changing index...
sa0 = strideA1; // stride for innermost loop
sa1 = strideA2; // stride for outermost loop
sb0 = strideB1;
sb1 = strideB2;
}
Progresses #2039
Description
This pull request:
blas/base/strsmRelated Issues
This pull request:
Questions
No.
Other
No.
Checklist
@stdlib-js/reviewers