add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502. #14508

Sacha0 · 2015-12-30T02:56:52Z

Followup to #14396 (comment). Subsumes #14505. ~~Fixes the right-division part of JuliaLang/LinearAlgebra.jl#295 (JuliaLang/LinearAlgebra.jl#295).~~ Update: Now subsumes #14504 as well, fixing JuliaLang/LinearAlgebra.jl#295 altogether.

This pull request provides generic methods, BLAS hooks, and tests for missing matrix-matrix and matrix-vector multiplication, left-division, and right-division operations involving one triangular matrix.

Concerning tests, the shadowing caveat mentioned in JuliaLang/LinearAlgebra.jl#295 / #14504 still stands. Right now I lean towards the simple fix of adding Complex{BigFloat} to the set of types tested in test/linalg/triangular.jl. Update: This pull request now expands the combinations of element types tested in linalg/triangular, fixing the shadowing issue and broadening the tests.

Question: Is the StridedVecOrMat part of the signature in

julia/base/linalg/triangular.jl

Line 401 in 6bf6f35

    
           A_rdiv_B!{T<:BlasFloat,S<:StridedMatrix}(A::StridedVecOrMat{T}, B::$t{T,S}) = BLAS.trsm!('R', $uploc, 'N', $isunitc, one(T), B.data, A)

correct? I would have thought the LHS could only be a StridedMatrix? Thanks!

tkelman · 2015-12-31T03:04:36Z

The nearly-repeated code here screams for some macro loops...

Sacha0 · 2015-12-31T03:37:23Z

The nearly-repeated code here screams for some macro loops...

Heh, check my last sentence in #14475 (comment) :). On the list.

andreasnoack · 2016-01-01T16:43:38Z

test/linalg/triangular.jl

@@ -278,13 +278,13 @@ for elty1 in (Float32, Float64, Complex64, Complex128, BigFloat, Int)
                @test full(A1 + A2) == full(A1) + full(A2)
                @test full(A1 - A2) == full(A1) - full(A2)

-                # Triangular-Triangular multiplication and division
-                elty1 != BigFloat && elty2 != BigFloat && @test_approx_eq full(A1*A2) full(A1)*full(A2)
+                # Triangular-Triangular multiplication and division # should transpose rather than only conjugate transpose variants be tested here?


I think we should test both versions here.

Tests expanded accordingly. Thanks!

andreasnoack · 2016-01-01T16:46:55Z

This is great.

Question: Is the StridedVecOrMat part of the signature in

I think it should be StridedMatrix

Sacha0 · 2016-01-01T17:53:43Z

I think it should be StridedMatrix

Fixed, thanks!

Sacha0 · 2016-01-01T20:57:00Z

The Travis failure seems unrelated? Failure on i686 with

signal (11): Segmentation fault
while loading /tmp/julia/share/julia/test/strings/basic.jl, in expression starting on line 248

…spectively. Fix generic Ac_mul_B! and A_mul_Bc! methods for complex triangular A and B respectively (see #14502). Add BLAS hooks for matrix-matrix At_mul_B! and A_mul_Bt! with triangular A and B respectively. Also add hooks for matrix-vector At_mul_B! and Ac_mul_B! with triangular A. Add tests for matrix-matrix and matrix-vector multiplications involving one transposed triangular matrix.

…div_Bc! methods for complex triangular B (see #14502). Add BLAS hook for matrix-matrix A_rdiv_Bt! with triangular B. Add tests for right-division by transposed triangular matrices.

…LAS hook for matrix-matrix and matrix-vector At_ldiv_B! with triangular A. Add tests for left-division by transposed triangular matrices.

…es test-shadowing of generic triangular methods by corresponding BLAS-calling methods for complex matrices (see #14502).

Sacha0 · 2016-01-11T02:08:36Z

A different unrelated Travis failure on x86_64 linux?

LoadError: start_watching (FileMonitor): no space left on device (ENOSPC)

Is this otherwise in shape to merge? Thanks!

tkelman · 2016-01-11T02:11:32Z

That's been happening occasionally since we switched to using docker workers on travis with deps caching, sorry. Restarted for now.

Sacha0 · 2016-01-11T02:17:30Z

Great, thanks!

andreasnoack · 2016-01-11T02:17:38Z

Looks good to me if @tkelman thinks the errors are unrelated.

Sacha0 · 2016-01-11T02:37:45Z

Same no space left on device error after the restart? CI does not seem to like this PR :).

add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502.

Sacha0 · 2016-01-12T03:15:11Z

Thanks for the review / restarts / merge!

Sacha0 mentioned this pull request Dec 30, 2015

add generic A(t|c)_ldiv_B! for triangular A #14505

Closed

Sacha0 force-pushed the trantriops branch from 40e130e to 9cd3f86 Compare December 30, 2015 21:09

Sacha0 mentioned this pull request Dec 30, 2015

Make generic Ac_mul_B! and A_mul_Bc! work for complex triangular A and B respectively #14504

Closed

Sacha0 force-pushed the trantriops branch from 9cd3f86 to 52bb784 Compare December 30, 2015 21:22

Sacha0 changed the title ~~add missing mul, ldiv, and rdiv variants involving one triangular matrix~~ add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502. Dec 30, 2015

Sacha0 mentioned this pull request Dec 30, 2015

fix promotion in bidiagonal backslash, At_ldiv_B, and Ac_ldiv_B. add missing bidiagonal matrix conversions. #14506

Merged

andreasnoack reviewed Jan 1, 2016
View reviewed changes

Sacha0 force-pushed the trantriops branch from 52bb784 to c0d5cf0 Compare January 1, 2016 18:00

kshyatt added the linear algebra Linear algebra label Jan 1, 2016

Sacha0 added 4 commits January 10, 2016 17:08

Add generic A_rdiv_Bt! methods for triangular B. Also fix generic A_r…

b9fc6e2

…div_Bc! methods for complex triangular B (see #14502). Add BLAS hook for matrix-matrix A_rdiv_Bt! with triangular B. Add tests for right-division by transposed triangular matrices.

Add generic At_ldiv_B! and Ac_ldiv_B! methods for triangular A. Add B…

52ba665

…LAS hook for matrix-matrix and matrix-vector At_ldiv_B! with triangular A. Add tests for left-division by transposed triangular matrices.

Expand combinations of element types tested in linalg/triangular. Fix…

009033c

…es test-shadowing of generic triangular methods by corresponding BLAS-calling methods for complex matrices (see #14502).

Sacha0 force-pushed the trantriops branch from c0d5cf0 to 009033c Compare January 11, 2016 01:22

tkelman added a commit that referenced this pull request Jan 11, 2016

Merge pull request #14508 from Sacha0/trantriops

95a01df

add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502.

tkelman merged commit 95a01df into JuliaLang:master Jan 11, 2016

Sacha0 deleted the trantriops branch January 12, 2016 03:15

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502. #14508

add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502. #14508

Uh oh!

Sacha0 commented Dec 30, 2015

Uh oh!

tkelman commented Dec 31, 2015

Uh oh!

Sacha0 commented Dec 31, 2015

Uh oh!

andreasnoack Jan 1, 2016

Uh oh!

Sacha0 Jan 1, 2016

Uh oh!

andreasnoack commented Jan 1, 2016

Uh oh!

Sacha0 commented Jan 1, 2016

Uh oh!

Sacha0 commented Jan 1, 2016

Uh oh!

Sacha0 commented Jan 11, 2016

Uh oh!

tkelman commented Jan 11, 2016

Uh oh!

Sacha0 commented Jan 11, 2016

Uh oh!

andreasnoack commented Jan 11, 2016

Uh oh!

Sacha0 commented Jan 11, 2016

Uh oh!

Sacha0 commented Jan 12, 2016

Uh oh!

Uh oh!

Uh oh!

add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502. #14508

add missing mul, ldiv, and rdiv variants involving one triangular matrix. also fix #14502. #14508

Uh oh!

Conversation

Sacha0 commented Dec 30, 2015

Uh oh!

tkelman commented Dec 31, 2015

Uh oh!

Sacha0 commented Dec 31, 2015

Uh oh!

andreasnoack Jan 1, 2016

Choose a reason for hiding this comment

Uh oh!

Sacha0 Jan 1, 2016

Choose a reason for hiding this comment

Uh oh!

andreasnoack commented Jan 1, 2016

Uh oh!

Sacha0 commented Jan 1, 2016

Uh oh!

Sacha0 commented Jan 1, 2016

Uh oh!

Sacha0 commented Jan 11, 2016

Uh oh!

tkelman commented Jan 11, 2016

Uh oh!

Sacha0 commented Jan 11, 2016

Uh oh!

andreasnoack commented Jan 11, 2016

Uh oh!

Sacha0 commented Jan 11, 2016

Uh oh!

Sacha0 commented Jan 12, 2016

Uh oh!

Uh oh!