On 04/05/17 22:28, Samuel Pitoiset wrote:
On 05/04/2017 02:13 PM, Samuel Pitoiset wrote:
On 05/04/2017 12:26 PM, Timothy Arceri wrote:
On 04/05/17 18:22, Samuel Pitoiset wrot
On 05/04/2017 02:36 AM, Timothy Arceri wrote:
This and the tgsi copy_propagation pass are very slow, I'd really
On 05/04/2017 02:13 PM, Samuel Pitoiset wrote:
On 05/04/2017 12:26 PM, Timothy Arceri wrote:
On 04/05/17 18:22, Samuel Pitoiset wrot
On 05/04/2017 02:36 AM, Timothy Arceri wrote:
This and the tgsi copy_propagation pass are very slow, I'd really
like it if we both didn't require the pass fo
On 05/04/2017 12:26 PM, Timothy Arceri wrote:
On 04/05/17 18:22, Samuel Pitoiset wrot
On 05/04/2017 02:36 AM, Timothy Arceri wrote:
This and the tgsi copy_propagation pass are very slow, I'd really
like it if we both didn't require the pass for things to work and
also didn't make it any slow
On 04/05/17 18:22, Samuel Pitoiset wrot
On 05/04/2017 02:36 AM, Timothy Arceri wrote:
This and the tgsi copy_propagation pass are very slow, I'd really like
it if we both didn't require the pass for things to work and also
didn't make it any slower. perf is showing these 2 passes at ~11%
durin
On 05/04/2017 02:36 AM, Timothy Arceri wrote:
This and the tgsi copy_propagation pass are very slow, I'd really like
it if we both didn't require the pass for things to work and also didn't
make it any slower. perf is showing these 2 passes at ~11% during Deus
Ex start-up (cold cache).
This
On 04/05/17 10:43, Ilia Mirkin wrote:
I looked into removing these passes a while back. A little detail here
is that DCE has to get run at least a little bit for correctness
reasons -- some (lazy) glsl -> tgsi
Is there anyway to improve this? I was actually looking at this a little
yesterday b
I looked into removing these passes a while back. A little detail here
is that DCE has to get run at least a little bit for correctness
reasons -- some (lazy) glsl -> tgsi code uses the dead_mask with the
assumption that DCE is run later.
Another observation was that while, indeed, those passes ta
This and the tgsi copy_propagation pass are very slow, I'd really like
it if we both didn't require the pass for things to work and also didn't
make it any slower. perf is showing these 2 passes at ~11% during Deus
Ex start-up (cold cache).
Can we not just add a DCE call to the glsl linker/com
On 05/03/2017 06:21 PM, Nicolai Hähnle wrote:
On 02.05.2017 16:43, Samuel Pitoiset wrote:
The TGSI DCE pass doesn't eliminate dead assignments like
MOV TEMP[0], TEMP[1] in presence of loops because it assumes
that the visitor doesn't emit dead code. This assumption is
actually wrong and this s
On 02.05.2017 16:43, Samuel Pitoiset wrote:
The TGSI DCE pass doesn't eliminate dead assignments like
MOV TEMP[0], TEMP[1] in presence of loops because it assumes
that the visitor doesn't emit dead code. This assumption is
actually wrong and this situation happens.
However, it appears that the m
The issue in Uningine Heaven is fixed.
Tested-by: Edmondo Tommasina
Thanks
edmondo
On Tue, May 2, 2017 at 4:43 PM, Samuel Pitoiset
wrote:
> The TGSI DCE pass doesn't eliminate dead assignments like
> MOV TEMP[0], TEMP[1] in presence of loops because it assumes
> that the visitor doesn't emit
The TGSI DCE pass doesn't eliminate dead assignments like
MOV TEMP[0], TEMP[1] in presence of loops because it assumes
that the visitor doesn't emit dead code. This assumption is
actually wrong and this situation happens.
However, it appears that the merge_registers() pass accidentally
takes care
12 matches
Mail list logo
