All on Juniper.
But anyway, I've got another solution, with the only drawback that I
don't really know what it does due to docs being lackluster/misleading
there :-).
But that would let us keep the ieee opcodes. And while I don't know what
it does, I suspect it's a better idea regardless ;-). hw s
Am 08.11.2017 um 07:18 schrieb Ilia Mirkin:
> tgsi_rsq appears to ignore the passed-in op and always puts in
> ALU_OP1_RECIPSQRT_CLAMPED anyways. It also sticks an absolute value on
> the RSQ() argument. This only happens for eg, not cayman. (Probably
> why only the rcp_clamped change appeared to b
Actually cayman gets half of it - it gets the abs, but not clamped. I
wonder what happens if you go the other way -- use the IEEE version of
the op for RSQ() (presumably you're not testing this on cayman).
On Wed, Nov 8, 2017 at 1:18 AM, Ilia Mirkin wrote:
> tgsi_rsq appears to ignore the passed-
tgsi_rsq appears to ignore the passed-in op and always puts in
ALU_OP1_RECIPSQRT_CLAMPED anyways. It also sticks an absolute value on
the RSQ() argument. This only happens for eg, not cayman. (Probably
why only the rcp_clamped change appeared to be necessary.)
This is odd though, because there's n
From: Roland Scheidegger
r600 already used the clamped versions, but for some reason this was
different to eg/cayman.
(Note that it has been different since essentially forever, 7 years, since
df62338c491f2cace1a48f99de78e83b5edd82fd in particular, which changed
this for r600 but not eg (cayman w
