276
61%

Ran 23 Jul 2017 04:25PM UTC

Jobs 2

Files 52

Run time 4s

Badge

Embed ▾

pending completion

Build # 276

Build Type

push

travis-ci

Committed by

benbarsdell

Commit Message

Add fast kernels for correlation and beamforming

- Adds custom kernels for particular cases of bfMatMul that
  significantly improve performance for these use-cases.
- The correlation kernel (a version of Cherk) is based on xGPU,
  and also includes a diagonal block optimization for improved
  performance at small N.
- The beamformer kernel (a version of Cgemm) uses a warp-sum
  approach and is optimized for small numbers of beams (<= ~12).
  It also supports mixed data/weights precisions (e.g., ci8 data and
  fp32 weights).
- Adds back-end plumbing to support batched linear algebra kernels.
- Modifies the matmul interface to allow either the lhs or rhs of
  symmetric products to be specified.
- Adds new tests for these kernels and fixes some old tests.

Run Details

2773 of 4253 relevant lines covered (65.2%)

1.3 hits per line

Jobs

ID	Job ID	Ran	Files	Coverage
1	276.1	23 Jul 2017 04:25PM UTC	0	65.2	Travis Job 276.1
2	276.2	23 Jul 2017 04:25PM UTC	0	65.2	Travis Job 276.2

Source Files on build 276

Detailed source file information is not available for this build.

ledatelescope / bifrost / 276
61%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 276

ledatelescope / bifrost / 276 61%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Jobs

Source Files on build 276

ledatelescope / bifrost / 276
61%

README BADGES
x