On 3/28/19 4:45 PM, Jason Ekstrand wrote:
On Thu, Mar 28, 2019 at 10:27 AM Samuel Pitoiset
mailto:samuel.pitoi...@gmail.com>> wrote:
Calling it after the first call of radv_optimize_nir() is worse.
Ugh Do we know why? I mean, it does emit the offset calculations
in a slightly diffe
On Thu, Mar 28, 2019 at 10:27 AM Samuel Pitoiset
wrote:
> Calling it after the first call of radv_optimize_nir() is worse.
>
Ugh Do we know why? I mean, it does emit the offset calculations in a
slightly different order but I wouldn't expect it to hurt this bad. :-/
> 27747 shaders in 143
Calling it after the first call of radv_optimize_nir() is worse.
27747 shaders in 14347 tests
Totals:
SGPRS: 1248039 -> 1248031 (-0.00 %)
VGPRS: 868360 -> 868772 (0.05 %)
Spilled SGPRs: 24108 -> 24134 (0.11 %)
Spilled VGPRs: 122 -> 122 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size:
On 3/28/19 3:08 PM, Jason Ekstrand wrote:
On Wed, Mar 27, 2019 at 4:13 AM Samuel Pitoiset
mailto:samuel.pitoi...@gmail.com>> wrote:
This helps few compute shaders, mostly for F12017.
27670 shaders in 14347 tests
Totals:
SGPRS: 1231173 -> 1231173 (0.00 %)
VGPRS: 866056 -> 8
On Thu, Mar 28, 2019 at 9:29 AM Jason Ekstrand wrote:
> I did some benchmarking of Assassin's Creed Odyssey today on ANV.
> Disabling the SSBO optimization possibilities in ANV that moving to derefs
> unlocks (I just moved the lowering super-early; moving back to index/offset
> would be insane) d
I did some benchmarking of Assassin's Creed Odyssey today on ANV.
Disabling the SSBO optimization possibilities in ANV that moving to derefs
unlocks (I just moved the lowering super-early; moving back to index/offset
would be insane) drops the perf of ACO by 20%. You want SSBO derefs. :D
On Thu,
On Wed, Mar 27, 2019 at 4:13 AM Samuel Pitoiset
wrote:
> This helps few compute shaders, mostly for F12017.
>
> 27670 shaders in 14347 tests
> Totals:
> SGPRS: 1231173 -> 1231173 (0.00 %)
> VGPRS: 866056 -> 865928 (-0.01 %)
> Spilled SGPRs: 24201 -> 24201 (0.00 %)
> Code Size: 46137040 -> 4614486
On 3/28/19 10:18 AM, Bas Nieuwenhuizen wrote:
R-b
Though not sure it really helps given code size increase?
I haven't benchmarked F12017 yet. The code size increases seems to be
LLVM related but that shouldn't matter much.
On Wed, Mar 27, 2019, 10:13 AM Samuel Pitoiset
mailto:samuel.pitoi..
R-b
Though not sure it really helps given code size increase?
On Wed, Mar 27, 2019, 10:13 AM Samuel Pitoiset
wrote:
> This helps few compute shaders, mostly for F12017.
>
> 27670 shaders in 14347 tests
> Totals:
> SGPRS: 1231173 -> 1231173 (0.00 %)
> VGPRS: 866056 -> 865928 (-0.01 %)
> Spilled
This helps few compute shaders, mostly for F12017.
27670 shaders in 14347 tests
Totals:
SGPRS: 1231173 -> 1231173 (0.00 %)
VGPRS: 866056 -> 865928 (-0.01 %)
Spilled SGPRs: 24201 -> 24201 (0.00 %)
Code Size: 46137040 -> 46144868 (0.02 %) bytes
Max Waves: 232287 -> 232302 (0.01 %)
Totals from affec
10 matches
Mail list logo
