Rearrange Lapack to be like BLAS. Implement cholesky decomposition

[btracey]

No description provided.

[kortschak]

You'll need to edit .travis.yaml to point to the correct location at line 42 s/clapack/cgo/.

[kortschak]

Prefix with "Package".

[btracey]

Done.

[btracey]

PTAL

[kortschak]

.travis.yaml needs to be updated still.

[btracey]

Travis was updated to swap clapack for cgo. It's still failing (obviously). Before it was giving a "pushd clapack" failure (or something like that). Now it's failing with

The command "go test -x -v ./..." exited with 2.
0.86s
$ diff <(gofmt -d .) <("")
/home/travis/build.sh: line 41: : command not found

There are a bunch of lapack.Job warnings as well (not sure what they mean). Is the problem that there are no go files in the cgo folder (just headers and a perl script?). Those have been there for a while, but I've never tried to integrate a c lapack installation.

[kortschak]

I'll have a look later today.

[kortschak]

This was needed by cgo/lapack.

[kortschak]

OK, figured out why it wasn't building: There are types defined in lapack/lapack.go that are required by lapack/cgo/lapack.go

[btracey]

Okay, cgo now compiles on my machine. I hacked the perl script in a way that's not pretty, but we'll need to update the script anyway at some point to use go generate, and have the right return types, etc.

[kortschak]

Are all the in returns in LAPACK 0=success !0=failure?

[kortschak]

Changes for genLapack.pl

diff --git a/cgo/genLapack.pl b/cgo/genLapack.pl
index 8c15042..07de5ef 100755
--- a/cgo/genLapack.pl
+++ b/cgo/genLapack.pl
@@ -57,12 +57,14 @@ const (
        rowMajor order = 101 + iota
        colMajor
 )
- 
+
 func init() {
          _ = lapack.Complex128(Lapack{})
          _ = lapack.Float64(Lapack{})
 }

+func isZero(ret C.int) bool { return ret == 0 }
+
 EOH

 $/ = undef;
@@ -110,11 +112,18 @@ our %typeConv = (
        "LAPACK_C_SELECT2" => "Select2Complex64",
        "LAPACK_Z_SELECT1" => "Select1Complex128",
        "LAPACK_Z_SELECT2" => "Select2Complex128",
-       "void" => ""
+       "void" => "",
+
+       "lapack_int_return_type" => "bool",
+       "lapack_int_return" => "isZero",
+       "float_return_type" => "float32",
+       "float_return" => "float32",
+       "double_return_type" => "float64",
+       "double_return" => "float64"
 );

 foreach my $line (@lines) {
-       process($line); 
+       process($line);
 }

 close($golapack);
@@ -162,14 +171,15 @@ sub processProto {
                $gofunc = ucfirst $func;
        }

-       my $GoRet = $typeConv{$ret};
+       my $GoRet = $typeConv{$ret."_return"};
+       my $GoRetType = $typeConv{$ret."_return_type"};
        my $complexType = $func;
        $complexType =~ s/.*_[isd]?([zc]).*/$1/;
        my ($params,$bp) = processParamToGo($func, $paramList, $complexType);
        if ($params eq "") {
                return
        }
-       print $golapack "func (Lapack) ".$gofunc."(".$params.") ".$GoRet."{\n";
+       print $golapack "func (Lapack) ".$gofunc."(".$params.") ".$GoRetType."{\n";
        print $golapack "\t";
        if ($ret ne 'void') {
                print $golapack "\n".$bp."\n"."return ".$GoRet."(";

[btracey]

Sorry, how do I apply that patch? I reverted back to the original genLapack.pl. I copied the text in the block above, saved it in cgo as patch.diff, and ran git apply -v patch.diff. I get the following output:

brendan:~/Documents/mygo/src/github.com/gonum/lapack/cgo$ git apply -v patch.diff 
Checking patch cgo/genLapack.pl...
error: while searching for:
       rowMajor order = 101 + iota
       colMajor
)

func init() {
         _ = lapack.Complex128(Lapack{})
         _ = lapack.Float64(Lapack{})
}

EOH

$/ = undef;

error: patch failed: cgo/genLapack.pl:57
error: cgo/genLapack.pl: patch does not apply

I don't understand the "error searching for" because those lines are in the file...

[btracey]

I'm not familiar with all of Lapack (learning as I go), but so far all of the returns are zero for success.

[btracey]

Okay, when I say "restored", I actually mean I copied and pasted in the old file. I wonder if that's part of the problem (where git is using a file index rather than just the characters?)

[kortschak]

Yeah, that will lead to a conflict.

[kortschak]

This last failure is due to removal of the test-coverage.sh file.

[btracey]

I thought we had removed all of those files. Do I remember incorrectly?

On May 8, 2015, at 4:47 AM, Dan Kortschak notifications@github.com wrote:

This last failure is due to removal of the test-coverage.sh file.

—
Reply to this email directly or view it on GitHub #11 (comment).

[kortschak]

Yes, you have. You should not have. The .travis.yaml calls the coverage script. This fails with 127 (mot found).

[btracey]

There we go. Sorry for making the process difficult.

[kortschak]

No worries, I'm learning about travis in the process - this is a good thing.

[btracey]

PTAL

[kortschak]

LGTM