
    sh$!                     &   S SK r S SKrS SKJr  S SKJr  SSKJrJr	  SSK
JrJrJrJr  SSKJrJrJrJrJr  SSKJr  S	S
KJrJrJrJrJrJr  \ R<                  " \5      r \RB                  RD                  r"\S 5       r#S r$S r%\" S\#SS9r&\" \RN                  S5      r(\" \RR                  S\"RR                  RT                  S9r+\	RX                  " \"RN                  5      SS.S j5       r-\	RX                  " \"RR                  5      S	S	SS.S j5       r.g)    N)counters)CKGemmTemplate   )irlowering)autotune_select_algorithmExternKernelChoiceSymbolicGridFnTritonTemplate)use_aten_gemm_kernelsuse_ck_gemm_templateuse_cpp_bmm_templateuse_cutlass_templateuse_triton_template)V   )_is_static_problemaddmm_epiloguemm_args
mm_configs
mm_optionsshould_fallback_to_atenc                6    U" XS   5      U" X#S   5      -  U S4$ )NBLOCK_MBLOCK_Nr    )bmnmetacdivs        n/Users/tiagomarins/Projetos/claudeai/copy_bank/venv/lib/python3.13/site-packages/torch/_inductor/kernel/bmm.pybmm_gridr#   %   s&    O$tAI'??AFF    c                 6    U S:  d  US:  d  US:  a  gX-  S:  $ )N   Ti   r   )r   r   ks      r"   _is_large_block_for_cpur(   *   s$    3w!c'QW55=r$   c                F    US:X  a  [        XUS[        S9$ [        XU5      $ )Ncpug      ?)scaleexclude)r   r(   )r   r   r'   device_types       r"   bmm_configsr.   1   s)    e!6MNNaAr$   bmma  
{{def_kernel("A", "B")}}
    M = {{size("A", -2)}}
    N = {{size("B", -1)}}
    K = {{size("A", -1)}}

    stride_aq = {{stride("A", 0)}}
    stride_am = {{stride("A", 1)}}
    stride_ak = {{stride("A", 2)}}

    stride_bq = {{stride("B", 0)}}
    stride_bk = {{stride("B", 1)}}
    stride_bn = {{stride("B", 2)}}

    # based on triton.ops.matmul
    pid = tl.program_id(0)
    grid_m = (M + BLOCK_M - 1) // BLOCK_M
    grid_n = (N + BLOCK_N - 1) // BLOCK_N

    # re-order program ID for better L2 performance
    width = GROUP_M * grid_n
    group_id = pid // width
    group_size = min(grid_m - group_id * GROUP_M, GROUP_M)
    pid_m = group_id * GROUP_M + (pid % group_size)
    pid_n = (pid % width) // (group_size)

    rm = pid_m * BLOCK_M + tl.arange(0, BLOCK_M)
    rn = pid_n * BLOCK_N + tl.arange(0, BLOCK_N)
    if (stride_am == 1 and stride_ak == M) or (stride_am == K and stride_ak == 1):
        ram = tl.max_contiguous(tl.multiple_of(rm % M, BLOCK_M), BLOCK_M)
    else:
        ram = rm % M
    if (stride_bk == 1 and stride_bn == K) or (stride_bk == N and stride_bn == 1):
        rbn = tl.max_contiguous(tl.multiple_of(rn % N, BLOCK_N), BLOCK_N)
    else:
        rbn = rn % N

    rk = tl.arange(0, BLOCK_K)

    idx_q = tl.program_id(1)  # batch dimension for BMM
    A = A + (ram[:, None] * stride_am + rk[None, :] * stride_ak + idx_q*stride_aq)
    B = B + (rk[:, None] * stride_bk + rbn[None, :] * stride_bn + idx_q*stride_bq)

    acc = tl.zeros((BLOCK_M, BLOCK_N), dtype=ACC_TYPE)
    for k in range(K, 0, -BLOCK_K):
        if EVEN_K:
            a = tl.load(A)
            b = tl.load(B)
        else:
            a = tl.load(A, mask=rk[None, :] < k, other=0.)
            b = tl.load(B, mask=rk[:, None] < k, other=0.)
        acc += tl.dot(a, b, allow_tf32=ALLOW_TF32)
        A += BLOCK_K * stride_ak
        B += BLOCK_K * stride_bk

    # rematerialize rm and rn to save registers
    rm = pid_m * BLOCK_M + tl.arange(0, BLOCK_M)
    rn = pid_n * BLOCK_N + tl.arange(0, BLOCK_N)
    idx_q = tl.program_id(1)  # batch dimension for BMM
    idx_m = rm[:, None]
    idx_n = rn[None, :]
    mask = (idx_m < M) & (idx_n < N)

    # inductor generates a suffix
    {{store_output(("idx_q", "idx_m", "idx_n"), "acc", "mask")}}
)namegridsourcezat::bmm_outzat::baddbmm_out)op_overloadlayoutc                  ^ [        S X4 5       5      (       Ga	  U R                  5       S   S:X  d  UR                  5       S   S:X  aW  [        R                  " U S5      n [        R                  " US5      n[        R                  " [        R
                  " X5      SS9$ S nS mU4S jnU" U 5      (       a/  [        R                  R                  R                  S	   nU" X5      n U" U5      (       a/  [        R                  R                  R                  S   nU" X5      n[        XUS
9u  pxpp[        S   SU SU SU	 3==   S-  ss'   [        R                  SUUU	U R                  5       UR                  5       U5        [        5       (       a  [         R#                  X4U5      /O/ n
[%        U5      (       aL  ['        XxU	[(        R*                  " U 5      S9 H)  n[,        R.                  " U
4X4US.[1        XXU5      D6  M+     [3        U5      u  pU(       a1  U(       a*  [5        X'X5      (       a  SSKJn  UR;                  XX/5        [=        X U5      (       a  SSKJ n  URC                  U
UX/5        [E        X'X5      (       a  [F        RH                  " XX/5        [K        U
5      (       a&  U
RM                  [         R#                  X4U5      5        [O        SXU/U5      $ )Nc              3   Z   #    U  H!  oR                  5       R                  S :H  v   M#     g7f)r*   N)
get_devicetype).0xs     r"   	<genexpr>tuned_bmm.<locals>.<genexpr>   s     
>A<<>%'s   )+r   r   )axisc                     [         R                  " U 5      (       d  g[         R                  " U SS9u  p[        U[         R                  5      $ )NTF)freeze)r   is_storage_and_layoutas_storage_and_layout
isinstanceFlexibleLayout)t_r5   s      r"   is_valid_to_require_contiguous1tuned_bmm.<locals>.is_valid_to_require_contiguous   s=    ++A..005AIAfb&7&788r$   c                     US   S:H  =(       a    U S   S:H  =(       d    US   U S   :  =(       d)    US   S:H  =(       a    U S   S:H  =(       d    US   U S   :  $ )Nr>   r   r   )sizesstridess     r"    is_preferred_layout_as_bmm_input3tuned_bmm.<locals>.is_preferred_layout_as_bmm_input   sf     q QeBi1n&PuRy8PU"+"Sb	Q(R'"+r:RUr$   c                    > UR                   S   R                  5       nUR                   S   R                  5       nT" X#5      (       d  [        R                  R                  U 5      n U $ )Nval)r    sizestrider   ExternKernelrequire_contiguous)rF   meta_trL   rM   rN   s       r"   may_require_contiguous)tuned_bmm.<locals>.may_require_contiguous   sU    KK&++-Ekk%(//1G3ECCOO66q9Hr$   r   r4   aten_mm_infoz	aten.bmm_rG   zPTuned aten.bmm: m=%s, n=%s, k=%s, mat1_dtype=%s, mat2_dtype=%s, output_layout=%sr-   input_nodesr5   )CUTLASS3xGemmTemplate)CppBmmTemplater/   )(allget_sizeL	unsqueezesum_mulr   graphcurrent_nodeargsr   r   loginfo	get_dtyper   aten_bmmbindr   r.   r   get_device_typebmm_templatemaybe_append_choicer   r   r   codegen.cuda.gemm_templater]   add_cutlass_gemm_choicesr   codegen.cpp_bmm_templater^   add_choicesr   r   add_ck_gemm_choicesr   appendr   )mat1mat2r5   rH   rW   	meta_mat1	meta_mat2r   r   r'   choicesconfigstatic_shape
is_nonzeror]   r^   rN   s                   @r"   	tuned_bmmr~      s   

>$
>>>==?1"dmmoa&8A&=;;tR(D;;tQ'D66!%%+!44	9	U	 *$//,,11!4I)$:D)$//,,11!4I)$:D")$V"DA!T ^y1QCq45:5HHZ			 8M7N7Nx}}d\623TVG6""!!r7I7I$7OPF,,!L Vf5	 Q  2&9L
';Fq'L'LF66wUF$//=""L	
 Fq,,**7TLIw''x}}d\6:;$UGD\6JJr$   )alphabetar5   c                B   [        XXS9u  pgppn [        S   SU SU SU 3==   S-  ss'   [        R                  SUUUUR	                  5       UR	                  5       U R	                  5       U5        [        5       (       a  [        R                  XU4XSUS9/O/ n	[        U5      (       af  [        XgU[        R                  " U5      S9 HC  n
[        R                  " U	4XU4US	.[        XXxU5      DS[        UR                   X45      S
.D6  ME     [#        SXX/U5      $ )Nr4   rY   zaten.baddbmm_rG   r   z\Tuned aten.baddbmm: m=%s, n=%s, k=%s, mat1_dtype=%s, mat2_dtype=%s, inp=%s, output_layout=%s)r   r   rZ   r[   )prefix_argsepilogue_fnbaddbmm)r   r   rh   ri   rj   r   aten_baddbmmrl   r   r.   r   rm   rn   ro   r   r   dtyper   )inprv   rw   r   r   r5   r   r   r'   rz   r{   s              r"   tuned_baddbmmr      s9   '.t3'N$A!T ^}QCq1QC89Q>9HHf				 !"" 
		Ct,f		MN 
 6""!!r7I7I$7OPF,, - Vf5	
 *6<<E Q %Yt9JFSSr$   )/loggingtorchtorch._dynamo.utilsr   7torch._inductor.codegen.rocm.ck_universal_gemm_templater    r   r   ra   select_algorithmr   r	   r
   r   utilsr   r   r   r   r   virtualizedr   	mm_commonr   r   r   r   r   r   	getLogger__name__rh   opsatenr#   r(   r.   rn   r/   rk   r   outr   register_loweringr~   r   r   r$   r"   <module>r      s     ( R       !yy~~ G G 		AEN eii7!	MM$$,,2B2B
 TXX$( SK SKl T\\",-Ad !T #!Tr$   