code review 145920043: math: fix amd64 Sincos CMPSD register handling

math: fix amd64 Sincos CMPSD register handling

CMPSD only populates half the SSE register, but we were using the entire register. This manifested itself as sin and cos being reversed in quadrants 1 and 2 in a program that used a cgo lib. I'm honestly not sure how this ever worked.

Fixes issue 8623.

[josharian, 2014-09-18 18:45:44]

Hello rsc@golang.org (cc: golang-codereviews@googlegroups.com),

I'd like you to review this change to
https://code.google.com/p/go

[iant, 2014-09-18 18:55:52]

https://codereview.appspot.com/145920043/diff/40001/src/math/sincos_amd64.s
File src/math/sincos_amd64.s (right):

https://codereview.appspot.com/145920043/diff/40001/src/math/sincos_amd64.s#n...
src/math/sincos_amd64.s:104: MOVHLPS X3, X3
I'm probably being an idiot but this doesn't make sense to me.  Don't you mean
MOVLHPS?

If all the tests pass with this instruction then I think we aren't using the
CMPSD result at all.  Or our tests suck.

Can you write a test, that does not use cgo, that passes with this instruction
and fails without it?

[josharian, 2014-09-18 18:58:59]

> I'm probably being an idiot but this doesn't make sense to me.  Don't you mean
> MOVLHPS?

Yikes. Nope, I'm the idiot. Sorry. Back to the drawing board.

[josharian, 2014-09-18 19:52:20]

*** Abandoned ***