mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-07-09 00:27:01 +00:00
b0f200713a
feat(ck-tile): add block-scale GEMM operators (aquant, bquant, abquant) (#8519) JIRA ID - AICK-1289 Motivation Adds three new block-scale quantized GEMM operators to the CK Tile Engine for FP8/BF8 inference workloads. Technical Details gemm_aquant: A-matrix quantized GEMM with per-row-group scale tensor [M, K/group_size_k] gemm_bquant: B-matrix quantized GEMM with per-column-group scale tensor [K/group_size_k, N] gemm_abquant: Both A and B quantized with independent group-scale tensors Each operator includes CMakeLists, Python instance builder with tier sampling, C++ benchmark/profiler with host reference verification, and config JSONs. Supporting changes to gemm_instance_builder.py, gemm_validation_utils.py, sampling infra, and the operation support matrix. Test Plan Build and run all three operators with fp8/bf8 on gfx942/gfx950 Verify correctness against CPU reference Verify CI config builds pass
23 lines
612 B
JSON
23 lines
612 B
JSON
{
|
|
"version": "1.0",
|
|
"description": "Op weights for Daily Tier budget allocation (RFC section 3.4)",
|
|
"weights": {
|
|
"gemm_universal": 0.0625,
|
|
"gemm_preshuffle": 0.0625,
|
|
"gemm_multi_d": 0.0625,
|
|
"grouped_gemm": 0.0625,
|
|
"gemm_streamk": 0.0625,
|
|
"batched_contraction": 0.0625,
|
|
"batched_gemm": 0.0625,
|
|
"gemm_multi_abd": 0.0625,
|
|
"mx_gemm": 0.0625,
|
|
"gemm_rowcolquant": 0.0625,
|
|
"gemm_tensor_quant": 0.0625,
|
|
"grouped_gemm_rowcolquant": 0.0625,
|
|
"grouped_gemm_tensorquant": 0.0625,
|
|
"gemm_aquant": 0.0625,
|
|
"gemm_bquant": 0.0625,
|
|
"gemm_abquant": 0.0625
|
|
}
|
|
}
|