On Thu, Nov 13, 2014 at 8:49 PM, Matt Turner <matts...@gmail.com> wrote:
> On Thu, Nov 13, 2014 at 4:28 PM, Kristian Høgsberg <k...@bitplanet.net> wrote:
>> The LOAD_PAYLOAD opcode can't saturate its sources, so skip
>> saturating MOVs. The register coalescing after lower_load_payload()
>> will clean up the extra MOVs.
>>
>> Signed-off-by: Kristian Høgsberg <k...@bitplanet.net>
>> ---
>> src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp | 6 +++++-
>> 1 file changed, 5 insertions(+), 1 deletion(-)
>>
>> diff --git a/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
>> b/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
>> index e1989cb..87ea9c2 100644
>> --- a/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
>> +++ b/src/mesa/drivers/dri/i965/brw_fs_copy_propagation.cpp
>> @@ -454,8 +454,12 @@ fs_visitor::try_constant_propagate(fs_inst *inst,
>> acp_entry *entry)
>> val.effective_width = inst->src[i].effective_width;
>>
>> switch (inst->opcode) {
>> - case BRW_OPCODE_MOV:
>> case SHADER_OPCODE_LOAD_PAYLOAD:
>> + /* LOAD_PAYLOAD can't sat its sources. */
>> + if (entry->saturate)
>> + break;
>> + /* Otherwise, fall through */
>> + case BRW_OPCODE_MOV:
>> inst->src[i] = val;
>> progress = true;
>> break;
>> --
>> 2.1.0
>
> This looks like the same patch as 01/14 in the previous series. I
> suggested a better approach there:
>
> At the beginning of fs_visitor::try_constant_propagate, if
> (entry->saturate) return false, or just saturate the argument and
> proceed. We don't want to be propagating the result of a mov.sat 4.0
> into anything without saturating the result first.
Fixed in v3.
Kristian
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
