Skip to content

Conversation

@deven-amd
Copy link
Contributor

@deven-amd deven-amd commented Apr 9, 2021

deven-amd and others added 4 commits April 9, 2021 17:27
The input buffer contents can be over-written by rocFFT for all of the real <---> complex transforms (R2C, C2R, D2Z, Z2D), even when transform explicitly provides a separate output buffer.

See following issues for more details

* ROCm/rocFFT#298
* #1150

This commit adds a change to copy the input buffer (for the above transforms) and use the input copy when calling the rocFFT api to do the transform
1. Some testcases were being skipped on the ROCm platform. Those tests are now passing and hence being enabled by this commit

2. For testcases that test the irfft transform, the input needs to be massaged on the ROCm platform. The ROCm irfft transform expects its input to be in the form that is output by the rfft transform.
…upstream_rocm_fft_fixes_210317

PiperOrigin-RevId: 365256946
Change-Id: I864143b70553fc2aeec8defb72dc30cb7442b10c
@google-cla google-cla bot added the cla: yes label Apr 9, 2021
@gbaned gbaned self-assigned this Apr 12, 2021
@gbaned gbaned added the size:M CL Change Size: Medium label Apr 12, 2021
@gbaned gbaned assigned mihaimaruseac and unassigned gbaned Apr 12, 2021
@gbaned gbaned requested a review from mihaimaruseac April 12, 2021 05:08
@mihaimaruseac mihaimaruseac merged commit 95d8865 into tensorflow:r2.5 Apr 22, 2021
@deven-amd deven-amd deleted the google_upstream_r25_port_pr_47937 branch May 12, 2021 23:47
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

cla: yes size:M CL Change Size: Medium

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants