Henrik RydgårdandClaude Opus 5.5 c4ce2e3635 SoftGPU: Use GERowSum4 for the transform's row sums ...

The four clip space rows of a vertex, the texture matrix's three, the world
normal's three, the vector to a light, and each row of the combined matrix
now take one GERowSum4 call. Same results (pspautotests, frametests and
the probe regression set unchanged).

GERowSum4 scales its result by adding to the exponent instead of
multiplying, and CrossSIMD gets a Vec4S32::Max for the largest lsb. About
68 M transforms/s against 26 for four scalar GERowSums on an M1 (a pure
NEON version with per-lane shifts did about 79); replaying the frametest
dumps gets only about 1% faster, up to 11% on Taiko no Tatsujin.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

2026-10-09 02:57:09 +02:00

..

Common

Merge pull request #22456 from hrydgard/spline-simd

2026-10-08 20:27:34 +02:00

D3D11

Software transform clipping: split clipped quads like the GE, cull beyond far first

2026-10-08 16:16:24 +02:00

Debugger

Merge pull request #22456 from hrydgard/spline-simd

2026-10-08 20:27:34 +02:00

GLES

Software transform clipping: split clipped quads like the GE, cull beyond far first

2026-10-08 16:16:24 +02:00

Software

SoftGPU: Use GERowSum4 for the transform's row sums

2026-10-09 02:57:09 +02:00

Vulkan

Software transform clipping: split clipped quads like the GE, cull beyond far first

2026-10-08 16:16:24 +02:00

CMakeLists.txt

GPU: Add the GE's depth buffer layout in VRAM

2026-10-08 18:35:48 +02:00

ge_constants.h

GPU: Fix draw engine buffer overruns and stale vertex data

2026-09-29 10:33:32 -06:00

GeConstants.cpp

Add the necessary logic to use in-shader CLUT lookups for regular textures

2026-06-11 12:42:16 +02:00

GeDisasm.cpp

Fix the ImGe debugger vertex list

2026-06-02 14:53:37 +02:00

GeDisasm.h

Build: Add some missing include guards.

2023-09-07 17:14:58 -07:00

GPU.cpp

GE debugger: Fix the stepping request handshake

2026-09-29 09:49:10 -06:00

GPU.h

Stop stacking duplicate video entries, and stop hashing video textures

2026-09-22 16:10:00 -06:00

GPU.vcxproj

GPU: Add the GE's depth buffer layout in VRAM

2026-10-08 18:35:48 +02:00

GPU.vcxproj.filters

GPU: Add the GE's depth buffer layout in VRAM

2026-10-08 18:35:48 +02:00

GPUCommon.cpp

GPU: Copy overlapping block transfer rows as the GE does

2026-10-09 02:03:31 +02:00

GPUCommon.h

GPU: Count the cycles of commands the hardware backends consume by hand

2026-10-09 02:03:31 +02:00

GPUCommonHW.cpp

GPU: Count the cycles of commands the hardware backends consume by hand

2026-10-09 02:03:31 +02:00

GPUCommonHW.h

Remove the "Software skinning" option. Now always on.

2026-07-14 17:02:57 +02:00

GPUDefinitions.h

Disable anisotropic filtering for all flat draws.

2026-06-08 15:28:58 +02:00

GPUState.cpp

GE dump playback: Restore old-layout contexts from old dumps

2026-10-08 18:37:32 +02:00

GPUState.h

GE dump playback: Restore old-layout contexts from old dumps

2026-10-08 18:37:32 +02:00

GPUStateSIMDUtil.h

Remove the cached UVScale in gstate_c. Conversion is cheap enough to do directly from gstate, no point in caching.

2026-07-03 13:17:35 +02:00

Math3D.cpp

Finish up this cleanup (move _M_SSE compat define out of Common.h)

2024-12-19 16:23:20 +01:00

Math3D.h

SoftGPU: Make arm64 and x86 render the GE probes identically

2026-10-02 15:09:43 -06:00