Repository navigation
Model L: D3D12 port of the halo input encoder (lev/input_encoder), byte for byte against its HIP kernel - #127
Merged
Merged
Conversation
…te for byte against its HIP kernel encoder.hlsl gets HIP's WG_ENC_VALUES_OUT switch (halo grid, unmirrored dither, x0 == x1 slot rule, halo map store, no projection); off by default. Refs #89 Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com>
Signed-off-by: phantoslayer21 <156617979+phantoslayer21@users.noreply.github.com> # Conflicts: # clean/l/runner/l_d3d12_ops_check.py
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Model L (#89): the D3D12 port of Model L's halo input encoder,
lev/input_encoder, byte for byte against its HIPkernel:
clean/kernels/encoder.hipbuilt withWG_ENC_VALUES_OUT=1asl_enc_values.co. The HIP kernel is merged in#115 and includes the width-17 fix.
Targets main (it extends the ops harness of #122 / #123). CI needs no GPU: nothing here runs without one.
clean/dx12/kernels/encoder.hlsl: HIP'sWG_ENC_VALUES_OUTswitch, off by default, plus the kernel's@wg-kernelline (co=l_enc_values.co, taglev). With the switch, as HIP:takes slot 0's depth neighbourhood and centre motion and keeps its own dither;
Buffer form only: an
#errorstops a build with direct textures or output-resolution motion. Model M's encodertext is unchanged outside the switch.
clean/l/runner/l_d3d12_ops_check.py: the encoder's cases, built from Model L kernels: encoder and decoder blocks, input linears, projections, coefficient step (HIP) #115'sl_enc_check.pyframes:40 x 24, 37 x 19, 64 x 32 (reset), 90 x 50, 17 x 17 and 17 x 24 (reset).
clean/reference/encoder.pyrandom_inputs, the blobenc0.values, andWgEncParamsby value.module l_enc_values.co lev) and the HIP kernel name(
input_encoder). The/of the launch name stays out of the case file names.Refs #89
Tested
Local, Windows, Radeon RX 9000 series card, against main's
l_enc_values.co:l_d3d12_ops_check.py --run --seeds 6 --kernels lev/input_encoder:lev/input_encoder: 36 of 36 cases identical (6 seeds x 6 frames; 4 outputs per case, 3,854,688 output bytes),including the width-17 frames;
swin_block_c96_nc: 36 of 36 identical (3,167,424 bytes);(HIP 0xA5, D3D12 0x5A).
build_dx12.py --isa(RGA): 4buffer_store_b128(the halo map, 2 per pixel) and the motion surfaces' 3buffer_store_b64+ 3buffer_store_b32, as HIP's; no matrix instructions.input_encoder,input_encoder_direct,emo/input_encoder_mo,emo/input_encoder_mo_direct): 4 of 4 byte-identical, main against thisbranch (DXIL SHA-256).
python tools/ci/check_line_endings.py: 0 problems. Public gates (secrets, personal data, licence, chunks) pass.Not tested