blksprs 1.0__py3-none-any.whl → 1.1__py3-none-any.whl

blksprs/ops/distribution.py

@@ -0,0 +1,362 @@
+import torch
+import triton
+from torch import Tensor
+from triton import language as tl
+
+from blksprs.utils.tools import get_triton_block_size
+from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_device, \
+    validate_sparsity, validate_dtype_int, validate_sparsity_block_size, validate_triton_block_size
+
+
+def gather(src: Tensor, sparsity_layout_src: Tensor, idx: Tensor, sparsity_layout_idx: Tensor,
+           sparsity_block_size: int, triton_block_size: int = None) -> Tensor:
+    """Applies a gather operation on a block-sparse tensor in compressed form.
+
+    Args:
+        src (Tensor): The source block-sparse tensor in compressed form to gather from.
+        sparsity_layout_src (Tensor): The sparsity layout of the source block-sparse tensor.
+        idx (Tensor): The block-sparse indices tensor in compressed form specifying how to gather from the source tensor.
+        sparsity_layout_idx (Tensor): The sparsity layout of the indices block-sparse tensor.
+        sparsity_block_size (int): The size of the sparsity blocks.
+        triton_block_size (int, optional): The block size to use for the triton kernel (default ``None``).
+
+    Returns:
+        Tensor: The result of the gather operation as a block-sparse tensor in compressed form.
+
+    """
+    validate_dimensions(src, idx)
+    validate_contiguous(src, idx)
+    validate_dtype_int(idx)
+    validate_device(src, idx)
+    validate_sparsity(sparsity_block_size, (src, sparsity_layout_src), (idx, sparsity_layout_idx))
+    validate_sparsity_block_size(sparsity_block_size, src, idx)
+    validate_triton_block_size(triton_block_size, sparsity_block_size)
+
+    sparsity_layout_x_flat = sparsity_layout_src.reshape(-1)
+    sparsity_reverse_lut_x = ((torch.cumsum(sparsity_layout_x_flat, dim=-1) - 1) *
+                              (sparsity_layout_x_flat == 1) -
+                              (1 * (sparsity_layout_x_flat == 0)))
+
+    sparsity_lut_i = torch.nonzero(sparsity_layout_idx).contiguous()
+
+    validate_contiguous(sparsity_layout_src, sparsity_reverse_lut_x,
+                        sparsity_layout_idx, sparsity_lut_i)
+
+    return _BlocksparseGather.apply(src, sparsity_layout_src, sparsity_reverse_lut_x,
+                                    idx, sparsity_layout_idx, sparsity_lut_i,
+                                    sparsity_block_size, triton_block_size)
+
+
+class _BlocksparseGather(torch.autograd.Function):
+
+    @staticmethod
+    def forward(ctx, x: Tensor, sparsity_layout_x: Tensor, sparsity_reverse_lut_x: Tensor,
+                i: Tensor, sparsity_layout_i: Tensor, sparsity_lut_i: Tensor,
+                sparsity_block_size: int, triton_block_size: int = None) -> Tensor:
+        output = torch.empty_like(i, dtype=x.dtype)
+
+        x_b, x_r, x_c = x.size()
+        x_b_s, x_r_s, x_c_s = x.stride()
+        s_l_x_b, s_l_x_r, s_l_x_c = sparsity_layout_x.size()
+        s_l_x_b_s, s_l_x_r_s, s_l_x_c_s = sparsity_layout_x.stride()
+        i_b, i_r, i_c = i.size()
+        i_b_s, i_r_s, i_c_s = i.stride()
+        s_lut_i_r, s_lut_i_c = sparsity_lut_i.size()
+        s_lut_i_r_s, s_lut_i_c_s = sparsity_lut_i.stride()
+        o_b, o_r, o_c = output.size()
+        o_b_s, o_r_s, o_c_s = output.stride()
+
+        if triton_block_size is None:
+            triton_block_size = get_triton_block_size(sparsity_block_size)
+
+        triton_grid = lambda meta: [o_b,
+                                    triton.cdiv(o_r, meta["TRITON_BLOCK_SIZE"]),
+                                    triton.cdiv(o_c, meta["TRITON_BLOCK_SIZE"])]
+
+        (_BlocksparseGather.kernel_blocksparse_gather[triton_grid]
+         (x,
+          x_b, x_b_s, x_r_s, x_c_s,
+          s_l_x_b, s_l_x_b_s, s_l_x_r_s, s_l_x_c_s,
+          sparsity_reverse_lut_x,
+          i,
+          i_b, i_b_s, i_r_s, i_c_s,
+          output,
+          o_b, o_b_s, o_r_s, o_c_s,
+          sparsity_lut_i, s_lut_i_r, s_lut_i_r_s, s_lut_i_c_s,
+          sparsity_block_size,
+          triton_block_size))
+
+        ctx.save_for_backward(sparsity_layout_x, i, sparsity_layout_i)
+        ctx.sparsity_block_size = sparsity_block_size
+        ctx.triton_block_size = triton_block_size
+
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        sparsity_layout_x, i, sparsity_layout_i = ctx.saved_tensors
+        sparsity_block_size = ctx.sparsity_block_size
+        triton_block_size = ctx.triton_block_size
+
+        return scatter_reduce(grad_output, sparsity_layout_i,
+                              i,
+                              sparsity_layout_x,
+                              sparsity_block_size,
+                              reduce_op="sum",
+                              triton_block_size=triton_block_size), None, None, None, None, None, None, None
+
+    @staticmethod
+    @triton.jit
+    def kernel_blocksparse_gather(x,
+                                  x_b, x_b_s, x_r_s, x_c_s,
+                                  s_l_x_b, s_l_x_b_s, s_l_x_r_s, s_l_x_c_s,
+                                  r_lut_x,
+                                  i,
+                                  i_b, i_b_s, i_r_s, i_c_s,
+                                  o,
+                                  o_b, o_b_s, o_r_s, o_c_s,
+                                  s_lut_o, s_lut_o_r, s_lut_o_r_s, s_lut_o_c_s,
+                                  sparsity_block_size,
+                                  TRITON_BLOCK_SIZE: tl.constexpr) -> None:
+        # Get triton block indices
+        pid_blk = tl.program_id(axis=0)
+        pid_row = tl.program_id(axis=1)
+        pid_col = tl.program_id(axis=2)
+
+        # Get position of current sparsity block consisting of its batch, row, and column index
+        spa_bat_o_idx = (pid_blk * s_lut_o_r_s + 0 * s_lut_o_c_s)
+        spa_bat_o_msk = (spa_bat_o_idx < s_lut_o_r * s_lut_o_r_s)
+        spa_bat_o = tl.load(s_lut_o + spa_bat_o_idx, mask=spa_bat_o_msk)
+
+        spa_row_o_idx = (pid_blk * s_lut_o_r_s + 1 * s_lut_o_c_s)
+        spa_row_o_msk = (spa_row_o_idx < s_lut_o_r * s_lut_o_r_s)
+        spa_row_o = tl.load(s_lut_o + spa_row_o_idx, mask=spa_row_o_msk)
+
+        # Load index values
+        blk_i_idx = ((pid_blk * i_b_s) +
+                     ((pid_row * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * i_r_s)[:, None] +
+                     ((pid_col * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * i_c_s)[None, :])
+        blk_i_msk = (blk_i_idx < i_b * i_b_s)
+        blk_i = tl.load(i + blk_i_idx, mask=blk_i_msk).to(tl.int32)
+
+        # Get positions of sparsity blocks
+        pos_spa_blk_x = blk_i // sparsity_block_size
+        pos_spa_col_x = blk_i % sparsity_block_size
+
+        # Load reverse sparsity indices for x
+        rev_idx_spa_x_idx = ((spa_bat_o * s_l_x_b_s) +
+                             (spa_row_o * s_l_x_r_s) +
+                             (pos_spa_blk_x * s_l_x_c_s))
+        rev_idx_spa_x_msk = (rev_idx_spa_x_idx < s_l_x_b * s_l_x_b_s)
+        rev_idx_spa_x = tl.load(r_lut_x + rev_idx_spa_x_idx, mask=rev_idx_spa_x_msk).to(tl.int32)
+
+        # Load x values
+        blk_x_idx = ((rev_idx_spa_x * x_b_s) +
+                     ((pid_row * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * x_r_s)[:, None] +
+                     (pos_spa_col_x * x_c_s))
+        blk_x_msk = (blk_x_idx < x_b * x_b_s)
+        blk_x = tl.load(x + blk_x_idx, mask=blk_x_msk)
+
+        # Store output
+        blk_o_idx = ((pid_blk * o_b_s) +
+                     ((pid_row * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * o_r_s)[:, None] +
+                     ((pid_col * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * o_c_s)[None, :])
+        blk_o_msk = (blk_o_idx < o_b * o_b_s)
+        tl.store(o + blk_o_idx, blk_x, mask=blk_o_msk)
+
+
+def scatter(src: Tensor, sparsity_layout_src: Tensor,
+            idx: Tensor,
+            sparsity_layout_tgt: Tensor,
+            sparsity_block_size: int, triton_block_size: int = None) -> Tensor:
+    """Wrapper for ``scatter_reduce`` with ``reduce_op="none"``.
+
+    """
+    return scatter_reduce(src, sparsity_layout_src,
+                          idx,
+                          sparsity_layout_tgt,
+                          sparsity_block_size,
+                          reduce_op="none", triton_block_size=triton_block_size)
+
+
+def scatter_reduce(src: Tensor, sparsity_layout_src: Tensor,
+                   idx: Tensor,
+                   sparsity_layout_tgt: Tensor,
+                   sparsity_block_size: int,
+                   reduce_op: str = "sum", triton_block_size: int = None) -> Tensor:
+    """Applies a scatter operation on a block-sparse tensor in compressed form.
+
+    Args:
+        src (Tensor): The source block-sparse tensor in compressed form to scatter from.
+        sparsity_layout_src (Tensor): The sparsity layout of the source block-sparse tensor.
+        idx (Tensor): The block-sparse indices tensor in compressed form specifying how to scatter to the target tensor.
+        sparsity_layout_tgt (Tensor): The sparsity layout of the target block-sparse tensor.
+        sparsity_block_size (int): The size of the sparsity blocks.
+        reduce_op (str, optional): The reduction operation to apply during the scatter operation (default ``"sum"``).
+            Supported operations are ``"none"`` and ``"sum"``.
+        triton_block_size (int, optional): The block size to use for the triton kernel (default ``None``).
+
+    Returns:
+        Tensor: The result of the scatter operation as a block-sparse tensor in compressed form.
+
+    """
+    validate_dimensions(src, idx)
+    validate_contiguous(src, idx)
+    validate_dtype_int(idx)
+    validate_device(src, idx)
+    validate_sparsity(sparsity_block_size, (src, sparsity_layout_src), (idx, sparsity_layout_src))
+    validate_sparsity_block_size(sparsity_block_size, src, idx)
+    validate_triton_block_size(triton_block_size, sparsity_block_size)
+
+    if reduce_op not in ["none", "sum"]:
+        raise ValueError(f"Reduction operation '{reduce_op}' is not supported")
+
+    sparsity_lut_x = torch.nonzero(sparsity_layout_src).contiguous()
+
+    sparsity_layout_o_flat = sparsity_layout_tgt.reshape(-1)
+    sparsity_reverse_lut_o = ((torch.cumsum(sparsity_layout_o_flat, dim=-1) - 1) *
+                              (sparsity_layout_o_flat == 1) -
+                              (1 * (sparsity_layout_o_flat == 0)))
+
+    n_sparse_blocks = torch.sum(sparsity_layout_tgt.to(torch.int)).item()
+
+    validate_contiguous(sparsity_layout_src, sparsity_lut_x,
+                        sparsity_layout_tgt, sparsity_reverse_lut_o)
+
+    return _BlocksparseScatterReduce.apply(src, sparsity_layout_src, sparsity_lut_x,
+                                           idx,
+                                           sparsity_layout_tgt, sparsity_reverse_lut_o,
+                                           sparsity_block_size, n_sparse_blocks,
+                                           reduce_op, triton_block_size)
+
+
+class _BlocksparseScatterReduce(torch.autograd.Function):
+
+    @staticmethod
+    def forward(ctx, x: Tensor, sparsity_layout_x: Tensor, sparsity_lut_x: Tensor,
+                i: Tensor,
+                sparsity_layout_o: Tensor, sparsity_reverse_lut_o: Tensor,
+                sparsity_block_size: int, n_sparse_blocks: int,
+                reduce_op: str, triton_block_size: int) -> Tensor:
+        output = torch.zeros(size=(n_sparse_blocks, sparsity_block_size, sparsity_block_size),
+                             dtype=x.dtype, device=x.device)
+
+        x_b, x_r, x_c = x.size()
+        x_b_s, x_r_s, x_c_s = x.stride()
+        s_lut_x_r, s_lut_x_c = sparsity_lut_x.size()
+        s_lut_x_r_s, s_lut_x_c_s = sparsity_lut_x.stride()
+        i_b, i_r, i_c = i.size()
+        i_b_s, i_r_s, i_c_s = i.stride()
+        o_b, o_r, o_c = output.size()
+        o_b_s, o_r_s, o_c_s = output.stride()
+        s_l_o_b, s_l_o_r, s_l_o_c = sparsity_layout_o.size()
+        s_l_o_b_s, s_l_o_r_s, s_l_o_c_s = sparsity_layout_o.stride()
+
+        if triton_block_size is None:
+            triton_block_size = get_triton_block_size(sparsity_block_size)
+
+        triton_grid = lambda meta: [x_b,
+                                    triton.cdiv(x_r, meta["TRITON_BLOCK_SIZE"]),
+                                    triton.cdiv(x_c, meta["TRITON_BLOCK_SIZE"])]
+
+        reduce_op_ind = 0
+        if reduce_op == "sum":
+            reduce_op_ind = 1
+
+        (_BlocksparseScatterReduce.kernel_blocksparse_scatter[triton_grid]
+         (x,
+          x_b, x_b_s, x_r_s, x_c_s,
+          sparsity_lut_x, s_lut_x_r, s_lut_x_r_s, s_lut_x_c_s,
+          i,
+          i_b, i_b_s, i_r_s, i_c_s,
+          output,
+          o_b, o_b_s, o_r_s, o_c_s,
+          s_l_o_b, s_l_o_b_s, s_l_o_r_s, s_l_o_c_s,
+          sparsity_reverse_lut_o,
+          reduce_op_ind,
+          sparsity_block_size,
+          triton_block_size))
+
+        ctx.save_for_backward(sparsity_layout_x, i, sparsity_layout_o)
+        ctx.sparsity_block_size = sparsity_block_size
+        ctx.reduce_op = reduce_op
+        ctx.triton_block_size = triton_block_size
+
+        return output
+
+    @staticmethod
+    def backward(ctx, grad_output):
+        sparsity_layout_x, i, sparsity_layout_o = ctx.saved_tensors
+        sparsity_block_size = ctx.sparsity_block_size
+        reduce_op = ctx.reduce_op
+        triton_block_size = ctx.triton_block_size
+
+        if reduce_op == "sum":
+            return gather(grad_output, sparsity_layout_o, i, sparsity_layout_x, sparsity_block_size,
+                          triton_block_size=triton_block_size), None, None, None, None, None, None, None, None, None
+        else:
+            raise ValueError(f"Reduction operation '{reduce_op}' does not support backward pass")
+
+    @staticmethod
+    @triton.jit
+    def kernel_blocksparse_scatter(x,
+                                   x_b, x_b_s, x_r_s, x_c_s,
+                                   s_lut_x, s_lut_x_r, s_lut_x_r_s, s_lut_x_c_s,
+                                   i,
+                                   i_b, i_b_s, i_r_s, i_c_s,
+                                   o,
+                                   o_b, o_b_s, o_r_s, o_c_s,
+                                   s_l_o_b, s_l_o_b_s, s_l_o_r_s, s_l_o_c_s,
+                                   r_lut_o,
+                                   reduce_op_ind,
+                                   sparsity_block_size,
+                                   TRITON_BLOCK_SIZE: tl.constexpr) -> None:
+        # Get triton block indices
+        pid_blk = tl.program_id(axis=0)
+        pid_row = tl.program_id(axis=1)
+        pid_col = tl.program_id(axis=2)
+
+        # Get position of current sparsity block consisting of its batch, row, and column index
+        spa_bat_x_idx = (pid_blk * s_lut_x_r_s + 0 * s_lut_x_c_s)
+        spa_bat_x_msk = (spa_bat_x_idx < s_lut_x_r * s_lut_x_r_s)
+        spa_bat_x = tl.load(s_lut_x + spa_bat_x_idx, mask=spa_bat_x_msk)
+
+        spa_row_x_idx = (pid_blk * s_lut_x_r_s + 1 * s_lut_x_c_s)
+        spa_row_x_msk = (spa_row_x_idx < s_lut_x_r * s_lut_x_r_s)
+        spa_row_x = tl.load(s_lut_x + spa_row_x_idx, mask=spa_row_x_msk)
+
+        # Load x values
+        blk_x_idx = ((pid_blk * x_b_s) +
+                     ((pid_row * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * x_r_s)[:, None] +
+                     ((pid_col * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * x_c_s)[None, :])
+        blk_x_msk = (blk_x_idx < x_b * x_b_s)
+        blk_x = tl.load(x + blk_x_idx, mask=blk_x_msk)
+
+        # Load index values
+        blk_i_idx = ((pid_blk * i_b_s) +
+                     ((pid_row * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * i_r_s)[:, None] +
+                     ((pid_col * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * i_c_s)[None, :])
+        blk_i_msk = (blk_i_idx < i_b * i_b_s)
+        blk_i = tl.load(i + blk_i_idx, mask=blk_i_msk).to(tl.int32)
+
+        # Get positions of sparsity blocks
+        pos_spa_blk_o = blk_i // sparsity_block_size
+        pos_spa_col_o = blk_i % sparsity_block_size
+
+        # Load reverse sparsity indices for o
+        rev_idx_spa_o_idx = ((spa_bat_x * s_l_o_b_s) +
+                             (spa_row_x * s_l_o_r_s) +
+                             (pos_spa_blk_o * s_l_o_c_s))
+        rev_idx_spa_o_msk = (rev_idx_spa_o_idx < s_l_o_b * s_l_o_b_s)
+        rev_idx_spa_o = tl.load(r_lut_o + rev_idx_spa_o_idx, mask=rev_idx_spa_o_msk).to(tl.int32)
+
+        # Store output
+        blk_o_idx = ((rev_idx_spa_o * o_b_s) +
+                     ((pid_row * TRITON_BLOCK_SIZE + tl.arange(0, TRITON_BLOCK_SIZE)) * o_r_s)[:, None] +
+                     (pos_spa_col_o * o_c_s))
+        blk_o_msk = (blk_o_idx < o_b * o_b_s)
+
+        if reduce_op_ind == 0:
+            tl.store(o + blk_o_idx, blk_x, mask=blk_o_msk)
+        elif reduce_op_ind == 1:
+            tl.atomic_add(o + blk_o_idx, blk_x, mask=blk_o_msk)

blksprs/ops/exp.py

@@ -1,25 +1,35 @@
 import torch
 import triton
-from triton import language as tl
 from torch import Tensor
+from triton import language as tl
 
 from blksprs.utils.tools import get_triton_block_size
-from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_dtype_float, validate_device
+from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_device, \
+    validate_sparsity_block_size, validate_triton_block_size
 
 
 def exp(x: Tensor, sparsity_block_size: int, triton_block_size: int = None) -> Tensor:
-    """Applies the element-wise exponential function to the input tensor.
+    """Applies the element-wise exponential function to a block-sparse tensor.
+
+    Note:
+        This operation does not consider sparse blocks, i.e., these will not be set to ``e^0``.
+        Consider this when converting back to tensors in regular form.
+
+    Args:
+        x (Tensor): A block-sparse tensor in compressed form.
+        sparsity_block_size (int): The size of the sparsity blocks.
+        triton_block_size (int): The block size to use for the triton kernel (default ``None``).
 
-        Returns a new tensor with the exponential of the elements of the input tensor.
+    Returns:
+        Tensor: The exponential function applied to all elements of the input tensor as a block-sparse tensor in
+            compressed form.
 
-        Note:
-            This operation does not consider sparse blocks, i.e., these will not be set to ``e^0``.
-            Consider this when converting back to dense tensors.
     """
     validate_dimensions(x)
     validate_contiguous(x)
-    validate_dtype_float(x)
     validate_device(x)
+    validate_sparsity_block_size(sparsity_block_size, x)
+    validate_triton_block_size(triton_block_size, sparsity_block_size)
 
     return _BlocksparseExp.apply(x, sparsity_block_size, triton_block_size)
 

blksprs/ops/{matmul_sss.py → matmul.py}

@@ -5,25 +5,39 @@ from triton import language as tl
 
 from blksprs.ops.transpose import transpose
 from blksprs.utils.tools import get_triton_block_size
-from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_dtype_float, validate_device, \
-    validate_sparsity
+from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_device, \
+    validate_sparsity, validate_sparsity_block_size, validate_triton_block_size
 
 
-def matmul_sss(x: Tensor, y: Tensor,
-               sparsity_layout_x: Tensor, sparsity_layout_y: Tensor, sparsity_layout_output: Tensor,
-               sparsity_block_size: int, triton_block_size: int = None) -> Tensor:
-    """Performs matrix multiplication between two blocksparse tensors.
+def matmul(x: Tensor, sparsity_layout_x: Tensor,
+           y: Tensor, sparsity_layout_y: Tensor,
+           sparsity_layout_output: Tensor,
+           sparsity_block_size: int, triton_block_size: int = None) -> Tensor:
+    """Performs matrix multiplication between two block-sparse tensors.
 
-    The desired sparsity layout of the output tensor is used to only calculate blocks that will be present in the output.
+    The sparsity layout of the output tensor is used to only calculate blocks that will be present in the output.
+
+    Args:
+        x (Tensor): A block-sparse tensor in compressed form.
+        y (Tensor): A block-sparse tensor in compressed form.
+        sparsity_layout_x (Tensor): The sparsity layout of the first block-sparse tensor.
+        sparsity_layout_y (Tensor): The sparsity layout of the second block-sparse tensor.
+        sparsity_layout_output (Tensor): The sparsity layout of the output tensor.
+        sparsity_block_size (int): The size of the sparsity blocks.
+        triton_block_size (int, optional): The block size to use for the triton kernel (default ``None``).
+
+    Returns:
+        Tensor: The result of the matrix multiplication as a block-sparse tensor in compressed form.
 
     """
     validate_dimensions(x, y)
     validate_contiguous(x, y)
-    validate_dtype_float(x, y)
     validate_device(x, y)
     validate_sparsity(sparsity_block_size, (x, sparsity_layout_x), (y, sparsity_layout_y))
     if sparsity_layout_x.size(-1) != sparsity_layout_y.size(-2):
         raise ValueError("Inner dimensions of tensors must match")
+    validate_sparsity_block_size(sparsity_block_size, x, y)
+    validate_triton_block_size(triton_block_size, sparsity_block_size)
 
     sparsity_layout_x_flat = sparsity_layout_x.reshape(-1)
     sparsity_reverse_lut_x = ((torch.cumsum(sparsity_layout_x_flat, dim=-1) - 1) *
@@ -98,10 +112,7 @@ class _BlocksparseMatmulSSS(torch.autograd.Function):
           sparsity_block_size,
           triton_block_size))
 
-        ctx.save_for_backward(x, y)
-        ctx.sparsity_layout_x = sparsity_layout_x
-        ctx.sparsity_layout_y = sparsity_layout_y
-        ctx.sparsity_layout_o = sparsity_layout_o
+        ctx.save_for_backward(x, sparsity_layout_x, y, sparsity_layout_y, sparsity_layout_o)
         ctx.sparsity_block_size = sparsity_block_size
         ctx.triton_block_size = triton_block_size
 
@@ -109,26 +120,17 @@ class _BlocksparseMatmulSSS(torch.autograd.Function):
 
     @staticmethod
     def backward(ctx, grad_output):
-        x, y = ctx.saved_tensors
-        sparsity_layout_x = ctx.sparsity_layout_x
-        sparsity_layout_y = ctx.sparsity_layout_y
-        sparsity_layout_o = ctx.sparsity_layout_o
+        x, sparsity_layout_x, y, sparsity_layout_y, sparsity_layout_o = ctx.saved_tensors
         sparsity_block_size = ctx.sparsity_block_size
         triton_block_size = ctx.triton_block_size
 
         x_t, sparsity_layout_x_t = transpose(x, sparsity_layout_x, sparsity_block_size, triton_block_size)
         y_t, sparsity_layout_y_t = transpose(y, sparsity_layout_y, sparsity_block_size, triton_block_size)
 
-        grad_x = matmul_sss(grad_output, y_t,
-                            sparsity_layout_o,
-                            sparsity_layout_y_t,
-                            sparsity_layout_x,
-                            sparsity_block_size, triton_block_size)
-        grad_y = matmul_sss(x_t, grad_output,
-                            sparsity_layout_x_t,
-                            sparsity_layout_o,
-                            sparsity_layout_y,
-                            sparsity_block_size, triton_block_size)
+        grad_x = matmul(grad_output, sparsity_layout_o, y_t, sparsity_layout_y_t, sparsity_layout_x,
+                        sparsity_block_size, triton_block_size)
+        grad_y = matmul(x_t, sparsity_layout_x_t, grad_output, sparsity_layout_o, sparsity_layout_y,
+                        sparsity_block_size, triton_block_size)
 
         return grad_x, grad_y, None, None, None, None, None, None, None, None, None
 

blksprs/ops/row_wise_sum.py

@@ -4,23 +4,39 @@ from torch import Tensor
 from triton import language as tl
 
 from blksprs.utils.tools import get_triton_block_size
-from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_dtype_float, validate_device
+from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_device, \
+    validate_sparsity, validate_sparsity_block_size, validate_triton_block_size
 
 
 def row_wise_sum(x: Tensor, sparsity_layout: Tensor, sparsity_block_size: int,
                  flag_slice_only: bool = False, triton_block_size: int = None) -> tuple[Tensor, Tensor]:
-    """Computes the row-wise sum of a blocksparse tensor.
+    """Computes the row-wise sum of a block-sparse tensor.
 
-    Returns a blocksparse tensor with only one block per row, where the first entry is the sum of the corresponding row.
+    Returns a block-sparse tensor in compressed form with only one block per row, where the first entry contains the sum
+        of the corresponding row.
 
     Note:
-        If ``flag_slice_only`` is set the output will be of shape ``[batch_size, row_size, 1]``.
+        If ``flag_slice_only`` is set the output will be of shape ``[x.size(0), x.size(1), 1]``.
+
+    Args:
+        x (Tensor): A block-sparse tensor in compressed form.
+        sparsity_layout (Tensor): The sparsity layout of the block-sparse tensor.
+        sparsity_block_size (int): The size of the sparsity blocks.
+        flag_slice_only (bool, optional): If set the output will be of shape ``[x.size(0), x.size(1), 1]``
+            (default ``False``).
+        triton_block_size (int): The block size to use for the triton kernel (default ``None``).
+
+    Returns:
+        tuple[Tensor, Tensor]: A tuple containing a block-sparse tensor in compressed form containing the row-wise sum
+            of the input and the sparsity layout of the output tensor.
 
     """
     validate_dimensions(x)
     validate_contiguous(x)
-    validate_dtype_float(x)
     validate_device(x)
+    validate_sparsity(sparsity_block_size, (x, sparsity_layout))
+    validate_sparsity_block_size(sparsity_block_size, x)
+    validate_triton_block_size(triton_block_size, sparsity_block_size)
 
     sparsity_lut = torch.nonzero(sparsity_layout).contiguous()
     sparsity_layout_flat = sparsity_layout.reshape(-1)

blksprs/ops/softmax.py

@@ -6,22 +6,37 @@ from triton import language as tl
 from blksprs.ops.exp import exp
 from blksprs.ops.row_wise_sum import row_wise_sum
 from blksprs.utils.tools import get_triton_block_size
-from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_dtype_float, validate_device
+from blksprs.utils.validation import validate_contiguous, validate_dimensions, validate_device, \
+    validate_sparsity, validate_sparsity_block_size, validate_triton_block_size
 
 
 def softmax(x: Tensor, sparsity_layout: Tensor, sparsity_block_size: int, triton_block_size: int = None) -> Tensor:
-    """Computes the softmax of a blocksparse tensor.
+    """Computes the softmax of a block-sparse tensor in compressed form.
 
     Note:
-        Sparse blocks are not considered for the calculation of the softmax, i.e., assumed to be ``-inf``.
+        Sparse blocks are not considered for the calculation of the softmax, i.e., all values are assumed to be ``-inf``.
+
+    Args:
+        x (Tensor): A block-sparse tensor in compressed form.
+        sparsity_layout (Tensor): The sparsity layout of the block-sparse tensor.
+        sparsity_block_size (int): The size of the sparsity blocks.
+        triton_block_size (int): The block size to use for the triton kernel (default ``None``).
+
+    Returns:
+        Tensor: The result of the softmax operation as a block-sparse tensor in compressed form.
 
     """
     validate_dimensions(x)
     validate_contiguous(x)
-    validate_dtype_float(x)
     validate_device(x)
-
-    max_val = torch.max(x).item()
+    validate_sparsity(sparsity_block_size, (x, sparsity_layout))
+    validate_sparsity_block_size(sparsity_block_size, x)
+    validate_triton_block_size(triton_block_size, sparsity_block_size)
+
+    if x.size(0) != 0:
+        max_val = torch.max(x).item()
+    else:
+        max_val = 0
     x_scaled = x - max_val
 
     sparsity_lut = torch.nonzero(sparsity_layout).contiguous()
@@ -83,9 +98,7 @@ class _BlocksparseSoftmax(torch.autograd.Function):
           triton_block_size))
 
         # Save for backward pass
-        ctx.save_for_backward(output)
-        ctx.sparsity_layout = sparsity_layout
-        ctx.sparsity_lut = sparsity_lut
+        ctx.save_for_backward(output, sparsity_layout, sparsity_lut)
         ctx.sparsity_block_size = sparsity_block_size
         ctx.triton_block_size = triton_block_size
 
@@ -93,9 +106,7 @@ class _BlocksparseSoftmax(torch.autograd.Function):
 
     @staticmethod
     def backward(ctx, grad_output):
-        o = ctx.saved_tensors[0]
-        sparsity_layout = ctx.sparsity_layout
-        sparsity_lut = ctx.sparsity_lut
+        o, sparsity_layout, sparsity_lut = ctx.saved_tensors
         sparsity_block_size = ctx.sparsity_block_size
         triton_block_size = ctx.triton_block_size
 

blksprs/ops/transpose.py

@@ -1,26 +1,37 @@
-from typing import Any
-
 import torch
 import triton
-from triton import language as tl
 from torch import Tensor
+from triton import language as tl
 
 from blksprs.utils.tools import get_triton_block_size
-from blksprs.utils.validation import validate_dimensions, validate_contiguous, validate_dtype_float, validate_device
+from blksprs.utils.validation import validate_dimensions, validate_contiguous, validate_device, \
+    validate_sparsity, validate_sparsity_block_size, validate_triton_block_size
 
 
 def transpose(x: Tensor, sparsity_layout: Tensor, sparsity_block_size: int, triton_block_size: int = None) -> (
         Tensor, Tensor):
-    """Transposes a blocksparse tensor.
+    """Transposes a block-sparse tensor in compressed form.
 
     Note:
          Returns the transposed tensor and the sparsity layout of the transposed tensor.
 
+    Args:
+        x (Tensor): A block-sparse tensor in compressed form.
+        sparsity_layout (Tensor): The sparsity layout of the block-sparse tensor.
+        sparsity_block_size (int): The size of the sparsity blocks.
+        triton_block_size (int): The block size to use for the triton kernel (default ``None``).
+
+    Returns:
+        Tensor: The transposed block-sparse tensor in compressed form.
+        Tensor: The sparsity layout of the transposed tensor.
+
     """
     validate_dimensions(x)
     validate_contiguous(x)
-    validate_dtype_float(x)
     validate_device(x)
+    validate_sparsity(sparsity_block_size, (x, sparsity_layout))
+    validate_sparsity_block_size(sparsity_block_size, x)
+    validate_triton_block_size(triton_block_size, sparsity_block_size)
 
     sparsity_layout_t = sparsity_layout.transpose(-1, -2).contiguous()
 
@@ -75,6 +86,7 @@ class _BlocksparseTranspose(torch.autograd.Function):
           triton_block_size))
 
         # Save for backward pass
+        ctx.save_for_backward(sparsity_layout)
         ctx.sparsity_layout = sparsity_layout
         ctx.sparsity_block_size = sparsity_block_size
         ctx.triton_block_size = triton_block_size
@@ -83,7 +95,7 @@ class _BlocksparseTranspose(torch.autograd.Function):
 
     @staticmethod
     def backward(ctx, grad_output):
-        sparsity_layout = ctx.sparsity_layout
+        sparsity_layout = ctx.saved_tensors[0]
         sparsity_block_size = ctx.sparsity_block_size
         triton_block_size = ctx.triton_block_size