Unknown W. Brackets
ae9d34370e
softgpu: Move wsum_recip out of the triangle loop.
...
Seems like a small benefit, but not seeing any issues from this.
Noticed by fp64.
2023-06-24 12:38:05 -07:00
Unknown W. Brackets
795de9b164
softgpu: Use SIMD for more Vec4 casts.
...
A number of these were falling back to some pretty terrible code.
Thanks to fp64 for noticing.
2023-06-24 12:36:44 -07:00
Unknown W. Brackets
76990aec70
Merge pull request #17609 from fp64/optimize-softgpu-tex-linear
...
softgpu: Optimize (bi-)linear texture filtering
2023-06-21 23:39:15 -07:00
fp64
159faaa2ec
softgpu: Optimize (bi-)linear texture filtering
...
Seeing as SampleLinearLevel is near the top in the profiler,
optimize actual bilinear filtering using SSE2. Solid win in the
synthetic benchmark (https://godbolt.org/z/fqh3xvbGx , also doubles
as correctness check), no visible difference in actual PPSSPP.
Note: profiler suggests that hot part of SampleLinearLevel is
elsewhere.
2023-06-21 20:02:34 +03:00
Henrik Rydgård
7cc8c6cea4
OSD: Add semantics, move the the OSD state to common (while keeping the renderer in the UI).
2023-06-20 14:40:46 +02:00
Unknown W. Brackets
efd8565ffe
Merge pull request #17592 from fp64/anymask-movemask
...
Use _mm_movemask_ps for AnyMask
2023-06-17 09:48:09 -07:00
fp64
ab85c46161
Use _mm_movemask_ps for AnyMask
...
Probably very minor speed improvement, but it's rather neat.
2023-06-17 01:05:02 -04:00
Henrik Rydgård
5b4fa06b00
Revert Dot33 on 32-bit x86 only. See #17584
2023-06-16 23:43:33 +02:00
Henrik Rydgård
6d4e5a0f3e
Merge pull request #17584 from fp64/sse2-dot33
...
Convert Dot33 to SSE2
2023-06-15 20:08:23 +02:00
Henrik Rydgård
def09bf575
Update the uvscale uniform a bit more conservatively on framebuffer changes
...
Plus fixes a few minor oversights
Fixes #17581 and possibly #17522
2023-06-15 11:57:30 +02:00
fp64
f0d844a5a3
Convert Dot33 to SSE2
...
Simpler, lower requirements, and doesn't seem to hurt speed. See #17571 .
2023-06-14 22:02:50 -04:00
Henrik Rydgård
6d8069dfd1
Vulkan: Remove the remains of the input attachment experiment
...
Haven't been using these for a while.
I've come to the conclusion here that I think it's better to try to
deal with the issues using safe workarounds like copies, instead of
relying on features with somewhat iffy driver support that are not
universal across APIs anyway.
2023-06-13 20:46:27 +02:00
Henrik Rydgård
df7bd89b7d
Division->shift. since it's a signed integer, gets rid of a cdq instruction.
2023-06-13 11:57:28 +02:00
Henrik Rydgård
0eb3702ecb
Then add the early-outs for NEON too.
2023-06-13 11:48:04 +02:00
Henrik Rydgård
9647872a09
Same for NEON, first the refactor...
2023-06-13 11:48:04 +02:00
Henrik Rydgård
77da36c03f
SSE addstrip: Add the early-outs.
2023-06-13 11:47:53 +02:00
Henrik Rydgård
39034586a4
SSE: Refactor AddStrip to prepare for early out
2023-06-13 11:45:59 +02:00
Henrik Rydgård
22632b82bd
Merge pull request #17565 from hrydgard/breakout-vcache-vulkan
...
Vulkan: Breakout the vertex cache logic from DoFlush()
2023-06-13 09:56:52 +02:00
Henrik Rydgård
963ca50ba7
Merge pull request #17567 from hrydgard/uvscale-as-argument
...
Pass uvScale in as a fourth argument to the vertex decoder
2023-06-13 09:49:31 +02:00
Henrik Rydgård
71a34d4ffc
Merge pull request #17569 from hrydgard/arm64dec-optimize-saved-regs
...
ARM64: Optimize saved registers in vertex decoder.
2023-06-13 09:49:08 +02:00
Unknown W. Brackets
a7fa37d114
softgpu: Use SIMD more for dot products.
2023-06-12 19:54:32 -07:00
Henrik Rydgård
cdcf3b272e
ARM64: Optimize saved registers in vertex decoder.
...
Simplify away some arrays with unused elements
2023-06-13 00:26:38 +02:00
Henrik Rydgård
4af6fac726
Nop-align the ARM and ARM64 loops too. Many CPUs benefit somewhat from hot loops being 16-byte aligned.
2023-06-13 00:05:48 +02:00
Henrik Rydgård
5ae9c9c64e
Merge pull request #17568 from hrydgard/extract-some-changes
...
Extract some minor changes from #17497
2023-06-12 23:38:14 +02:00
Henrik Rydgård
c4e44d66b0
x86/x64: Nop-align the main loop of vertex decoder loops
2023-06-12 20:39:39 +02:00
