Function grid

src/fused_mm_sampling/persistent_matmul.py:267–270

(META)

    c_desc = TensorDescriptor.from_tensor(c, dummy_block)

    def grid(META):
        BLOCK_M = META["BLOCK_SIZE_M"]
        BLOCK_N = META["BLOCK_SIZE_N"]
        return (triton.cdiv(M, BLOCK_M) * triton.cdiv(N, BLOCK_N),)

    matmul_kernel_tma[grid](
        a_desc,