Conversation
joaosaffran
left a comment
joaosaffran
left a comment
There was a problem hiding this comment.
The testing logic looks fine. Left some comments regarding config changes. We need to specify some ULP level when working with floating types.
| Actual: FOut4 | ||
| Expected: FExpectedOut4 | ||
| - Result: FExpectedOut5 | ||
| Rule: BufferExact |
There was a problem hiding this comment.
I think we should have an small ULP here, like 1, for the float outputs
| ... | ||
| #--- end | ||
|
|
||
| # Bug: https://github.com/llvm/llvm-project/issues/186151 |
There was a problem hiding this comment.
This bug is regarding WavePrefixSum, is the same bug happening in here? If that is the case we should have a different bug anyways
| # Bug: https://github.com/llvm/offload-test-suite/issues/351 | ||
| # XFAIL: Metal | ||
|
|
||
| # Bug: https://github.com/llvm/offload-test-suite/issues/961 |
There was a problem hiding this comment.
Another WavePrefixSum specific bug
| - Name: In | ||
| Format: Float16 | ||
| Stride: 8 | ||
| Data: [ 0x4000, 0x4200, 0x4500, 0x4700 ] |
There was a problem hiding this comment.
Can we get a comment bellow showing what those numbers are please?
| Data: [ 0x3c00, 0x3c00, 0x3c00, 0x3c00, 0x4000, 0x4200, 0x4500, 0x4700, 0x4400, 0x4880, 0x4e40, 0x5220, 0x4800, 0x4ec0, 0x57d0, 0x5d5c ] | ||
| Results: | ||
| - Result: ExpectedOut1 | ||
| Rule: BufferExact |
There was a problem hiding this comment.
We definitely need ULP here. It should be higher than the float case, like 2 since I think the offload test expects integers in here.
|
Also, can we add the |
bob80905
added
the
test-all
3 participants
Adds WavePrefixProduct tests.
Based on WavePrefixSum, except the first lane returns 1 instead of 0.
Fixes #914