Commit Graph

14 Commits

Author SHA1 Message Date
nmzik defb2f11d3 Share median classifications and preserve RDNA2 signed-zero selection 2026-10-19 02:34:25 +02:00
AN0DIN 83db2efec3 shader: pack v_cvt_pkrtz_f16_f32 with the native half conversion (GTA V) (#1520)
* shader: pack v_cvt_pkrtz_f16_f32 with the native half conversion

The round-toward-zero pack was emulated with integer operations: about 20 to 30 operations
per half. GTA V uses it in 613 of its 656 pixel shaders, about 16 times in each G-buffer
pixel shader.

The pair is now packed with PackHalf2x16, unpacked again, and each half whose magnitude
exceeds its input's takes one unit off. Half signs and magnitudes are separate, so that is
the next value toward zero, whatever rounding the native conversion uses as long as it
returns one of the two neighbours. Same bits as before for every finite input and infinity;
a NaN stays a NaN. Checked against the old code on 2 million values on the CPU, and on an
Apple M5 GPU through Metal (400 000 values, subnormals included, safe and fast math).

GTA V on an Apple M5 (MoltenVK), 1080p, fixed script camera, three places, A B B A between
launches: -1.7 to -2.8 ms per frame (+4 to +6 %).

Scope: generic
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

* shader: remove half-pack commentary and unused shift helper

---------

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: nmzik <Nmzik@mail.ru>
2026-10-16 02:56:35 +02:00
nmzik 393257eaae shader: classify float NaNs with one magnitude comparison 2026-10-09 02:54:21 +02:00
nmzik ee4dd90649 shader: emit native integer operations and remove signedness casts 2026-10-09 02:54:21 +02:00
nmzik 7a0e1dafd7 shader: classify native half values with shared operand decoding 2026-10-05 18:39:31 +04:00
nmzik 0791f921f6 UFC 5 in-game (#883)
* shader: release descriptor dependencies

* shader: simplify shared-exit control flow

* shader: fix scalar memory loads

* shader: fix LDS counter bounds

* renderer: skip inactive pixel shaders

* shader: respect FP32 mode in comparisons

* renderer: fix sampler border colors

* shader: resolve descriptors from control flow

* shader: resolve descriptors before resource planning

* shader: handle scalar guards for image bounds

* shader: handle reduced guards for image loops

* shader: fix wave guards and material indices

* shader: preserve scalar buffer offset alignment and carry

* shader: prove masked loop bounds across iteration transitions

* shader: preserve source and IR before resource tracking failures

* shader: track material keys selected from active lanes

* shader: remove obsolete post-dominator analysis

* shader: follow bounds guards through control-flow joins
2026-09-28 16:03:05 +04:00
nmzik 1208d22547 Fix Windows Clang SPIR-V serialization with typed operands 2026-09-12 13:08:23 +02:00
nmzik 20e57994cf shader: reuse native emitters and simplify memory dispatch 2026-09-12 13:42:15 +04:00
nmzik 1d1f00bdef shader: use Khronos definitions and simplify compiler ownership 2026-09-12 13:42:15 +04:00
nmzik 9f49ac8249 shader: reuse SPIR-V helpers and memory resources 2026-09-08 21:26:59 +04:00
nmzik 9241fdc32a shader: align float semantics with RDNA2 default mode 2026-08-22 03:16:45 +02:00
nmzik 6e9f162bd4 refactor(shader): compact paired 64-bit lowering 2026-08-21 03:32:47 +02:00
nmzik d34e30d892 refactor(shader): emit SPIR-V declarations on demand 2026-08-21 03:32:47 +02:00
nmzik 01275fb211 shader_recompiler: rename/clean up typed SSA SPIR-V emitter 2026-08-18 03:53:45 +02:00