Composite GEMM: K-loop accumulator residency, pinned operands, sweep + deck

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-13 15:00:41 -07:00
parent 5accd98171
commit 83ea97b05f
11 changed files with 4219 additions and 51 deletions
@@ -34,6 +34,7 @@ class TensorHandle:
    nbytes: int                      # total byte size
    data: object = None              # reserved for validate mode
    space: str = "tcm"               # MemoryStore space ("tcm" | "hbm" | "sram")
+    pinned: bool = False             # operand already DMA-staged in TCM (via tl.load)


@dataclass(frozen=True)