[ET-VK] Layout-flexible impl of quantized binary

[SS-JIA]

Stack from ghstack (oldest at bottom):

• [ET-VK][profiling] Add additional profiling blocks #17394
• [ET-VK][qconv2d][ez] Don't use im2col path for general convs #17393
• [ET-VK] Add nchw_to_int8x4_buffer shader for prepacking int8 staging data #17392
• -> [ET-VK] Layout-flexible impl of quantized binary #17391
• [ET-VK][ez] Make q8ta_conv2d use 4C1W layout #17390
• [ET-VK][ez] Add AOT support for PackedInt8_4C1W dtype #17389
• [ET-VK] Migrate to use new q8ta_conv2d ops #17388
• [ET-VK][qconv] Dynamically select between im2col path and general path #17387
• [ET-VK][qconv][ez] Make q8ta_im2col shader support stride_w != 1 #17386
• Back out "[Diff Train][pytorch/executorch] Apply fixup patch to fbsource" #17399

This refactors the quantized binary add operator to support all PackedInt8 memory layouts (4W, 4C, 4W4C, 4H4W, 4C1W) instead of being hardcoded to 4W4C. The shader is rewritten to use the block indexing framework (BlockConfig, block_int8x4_load/store) and BufferMetadata for layout-agnostic tensor access, replacing the previous linear dispatch that assumed 4W4C ordering.

Key changes:

• Renames shader from binary_q8ta_q8ta_q8to to q8ta_binary, and op from add_q8ta_q8ta_q8to to q8ta_add
• Shader now uses contiguous_block_idx_to_tensor4d_idx_with_block_config for dispatch and generated load/store functions for layout-flexible int8x4 access
• C++ dispatch uses pick_linear_global_wg_with_block_config and passes BufferMetadata UBOs for output and both inputs, plus hashed_layout specialization constants
• Moves the test operator into a separate TestQ8taBinary.cpp file that parameterizes on GPUMemoryLayout, testing all 5 layouts
• Updates op_registry to accept PACKED_INT8_BUFFER (all layouts) instead of just PACKED_INT8_4W4C_BUFFER

This diff was authored with Claude.

Differential Revision: D93000170

[pytorch-bot]

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/17391

• 📄 Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

❌ 1 New Failure, 4 Unrelated Failures

As of commit c2ca434 with merge base 964c565 ():

NEW FAILURE - The following job has failed:

• pull / test-vulkan-operators-linux / linux-job (gh)
  RuntimeError: Command docker exec -t 38cdb8a511e86e769a15d3c7abd75dbd514a910af1e786152c7a4728718f3755 /exec failed with exit code 1

FLAKY - The following jobs failed but were likely due to flakiness present on trunk:

• pull / test-custom-ops-linux / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• pull / test-models-linux-basic (vit, xnnpack-quantization-delegation, buck2, linux.2xlarge, executorch-u... / linux-job (gh) (detected as infra flaky with no log or failing log classifier)
• pull / unittest-nxp-neutron / linux-job (gh) (detected as infra flaky with no log or failing log classifier)

BROKEN TRUNK - The following job failed but were present on the merge base:

👉 Rebase onto the `viable/strict` branch to avoid these failures

• pull / test-samsung-models-linux / linux-job (gh) (trunk failure)
  ##[error]The operation was canceled.

This comment was automatically generated by Dr. CI and updates every 15 minutes.

[github-actions]

This PR needs a release notes: label

If your change should be included in the release notes (i.e. would users of this library care about this change?), please use a label starting with release notes:. This helps us keep track and include your important work in the next release notes.

To add a label, you can comment to pytorchbot, for example
@pytorchbot label "release notes: none"

For more information, see
https://github.com/pytorch/pytorch/wiki/PyTorch-AutoLabel-Bot#why-categorize-for-release-notes-and-how-does-it-work.