0888 KRGB and 8888 ARGB, still went through the scalar per-pixel writer.
One packed pixel is exactly one VRAM word, so they fit the same 8-word
interleaved store as the 16-bit path, 8 pixels per iteration. The store
is factored out of storeFramebuffer16Pixels() so both use it.
888 (24-bit packed) stays scalar: its pixels straddle VRAM words, so the
store pattern only repeats every 4 pixels / 3 words.
The address walk is the same as writeFramebufferLW(), so the output is
identical to the scalar path, including when the padding isn't a
multiple of 4. A harness writing 7,056 frames through both paths (both
channel orders, modes 4 to 6, 6 frame sizes, 7 destination addresses
including bank and VRAM edges, 7 line strides, 4 clip windows) gives
byte-identical VRAM. The same algorithm measured about 2x faster on a
640x480 8888 frame (0.213 ms to 0.105 ms) on Apple Silicon.
tests/src/rend/FramebufferWriteTest.cpp checks both 32-bit modes against
an independent reference (one pvr_write32p per pixel, the scalar walk)
over 3,600 cases: widths around the 8-pixel vector width, padding
including values that aren't a multiple of 4, destinations at the 32-bit
bank boundary and the end of VRAM, and clip windows, for both channel
orders. All of VRAM is compared, so a store past the frame fails too. On
arm64 it exercises the NEON path; elsewhere the scalar one. It fails when
the padding computation is deliberately broken.
Other architectures are unchanged.
Co-authored-by: Claude Opus 5.5 <[email protected]>
[ci skip]
xbrz::scale() only implements factors 2 to 6. Any other value hits an
assert that's compiled out in release builds, and scale() returns
without writing the destination buffer. The texture is then uploaded
from uninitialized memory at the requested size.
rend.TextureUpscale2 can be set to 8 or 16 through emu.cfg or Lua.
Clamp the factor once and use it for the buffer size, the scale call
and the upscaled dimensions so they always agree.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Add NEON framebuffer conversion and VRAM write paths, and validate the physical VRAM footprint before vector loads so both 8 MB and 16 MB configurations remain in bounds.
[ci skip]
Catch/rethrow xbyak exceptions in arm7 and sh4 dynarecs. Fixes
MINIDUMP-8P9
naomi: code to generate random main board serial id (not used yet)
Don't log warning for sh4 trapa op.
texcache: get rid of unused FBLineWriter argument.
Fix regression with split-screen multi render in Quake 3. Caused by
aef3d2d337.
Move duplicated code into transform_matrix. Handle framebuffer
dimensions/clipping scaling depending on render mode (Full FB emu, copy
RTT to vram).
Handle RTT and screen render similarly when possible.
[ci skip]
Fix framebuffer clipping when scale>1 (Stretched PAL). Issue #728
Get rid of ScissorMatrix.
Don't scale framebuffer clipping rect in full FB emu.
Refactor rend_context. Use Rect and ivec2.
Only one palette is supported in the shader, so convert area 1 palette
textures (2-volume mode) on the cpu and handle them normally in the
shader.
Issue #1528
Some textures are frequently updated because of their location in vram,
close to a hotspot. This causes the custom texture to never be loaded.
This change will skip the unneeded texture update, allowing the custom
tex to replace it when available.
Issue #1812
Force palette update when changing directx color order.
Add 10k extra depth scale to Capcom vs SNK 2000 Pro and cvs.zip to fix
winner character portraits not appearing during ending.
Issue #1546
The calculated size for vq textures was half the correct value, which
affects the texture hash. Calculate the old hash and use it as fallback
for backward-compatibility.
Issue #1291
Make the texture data start address really point to max mipmap data, no
more codebook skipping.
gl, vk, dx11, dx9: implement ordered dithering in GLSL/HLSL
refactor framebuffer and texture writes to vram
don't expand color data when reading planar textures
Ignore StrideSel if 0 during texture update and use width instead.
Return false from BaseTextureCacheData::Update() if texture is invalid.
Make sure valid textures have non-null height and size.
gl,vk,dx9,dx11: ignore invalid textures
Fixes MINIDUMP-GP
* cmake: use c++17
* Use std::size
* Use std::make_unique
* Use std::clamp
* Use structured binding
* Use [[fallthrough]]
* Use enable_if_t/is_enum_v/is_integral_v/is_same_v
* Use if constexpr
* Use try_emplace
* Use auto for iterators
* Use inline variables
Part of PalSelect was used as cache key for palette textures
PrintTexture fix
egl,wgl,xgl: No need for depth/stencil surface
gl: non-functional refactoring
Store FT_W_SOF1 and FT_W_CTRL in render context since they can be
modified before render end. New tentative fix for vf4 water reflection
flashing black during netplay.
Refactor texture cache to use constructors. RTT code moved from
renderers.
Partial revert of ddcd3861d3
Joe Mattiello