[vulkan_api][ops] Mm, Pool, Upsample #47063
Conversation
[ghstack-poisoned]
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
| mat2_arg.sizes()[1], | ||
| mat1_arg.sizes()[0], | ||
| 1u, | ||
| mat1_arg.sizes()[1], |
There was a problem hiding this comment.
Should be the last argument? (i.e. corresponds to K in shader Block)
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
💊 CI failures summary and remediationsAs of commit 5375bb6 (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 25 times. |
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
ghstack-source-id: c939b3b5c264421288b3c52a0720ac2f62d585d2 Pull Request resolved: #47063
|
|
||
| vTensor v_output{ | ||
| context, | ||
| self.sizes(), |
There was a problem hiding this comment.
I changed this to
{mat1.sizes()[0], mat2.sizes()[1]},
and the addmm test is now passing on my laptop
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
| command_buffer.begin(); | ||
| { | ||
| const float scale_x = compute_scales_value<float>(scales_w, input_arg.sizes()[3], output_sizes[2]); | ||
| const float scale_y = compute_scales_value<float>(scales_h, input_arg.sizes()[3], output_sizes[2]); |
There was a problem hiding this comment.
I think the scales are being computed wrong here (mismatched indices), which is causing the test to fail.
Perhaps should be something like
scale_x = input_arg.sizes()[2]/output_sizes[0]
scale_y = input_arg.sizes()[3]/output_sizes[1]
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
|
|
||
| vTensor v_output{ | ||
| context, | ||
| output_sizes, |
There was a problem hiding this comment.
Since output_sizes only has 2 elements, I also think this should be something like
{input_arg.sizes()[0], input_arg.sizes()[1], output_sizes[0], output_sizes[1]
| input_arg.sizes()[3], | ||
| input_arg.sizes()[2], | ||
| output_sizes[3], | ||
| output_sizes[2], |
There was a problem hiding this comment.
Should be output_sizes[1], output_sizes[0] here as well
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
ghstack-source-id: 13da256a2f788240980fc58a3eceb4612cccc09d Pull Request resolved: #47063
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
ghstack-source-id: fee359030ed1bfab2eb613fc7ce3b4d41e0da119 Pull Request resolved: #47063
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
ghstack-source-id: 3d343b7998d19125b402dfb285544c8578a3d910 Pull Request resolved: #47063
Differential Revision: [D24624610](https://our.internmc.facebook.com/intern/diff/D24624610) [ghstack-poisoned]
ghstack-source-id: 6a59a73d63c48e9c406e669fab7be8af6dfc1ca0 Pull Request resolved: #47063
|
@IvanKobzarev merged this pull request in 2cff3bb. |
Stack from ghstack:
Differential Revision: D24624610