Cảm ơn bạn đã theo dõi Google I/O. Xem tất cả các phiên theo yêu cầu Xem theo yêu cầu

Ngữ nghĩa hoạt động

Phần sau đây mô tả ngữ nghĩa của các hoạt động được xác định trong giao diện XlaBuilder . Thông thường, các thao tác này ánh xạ một đối một với các thao tác được xác định trong giao diện RPC trong xla_data.proto .

Một lưu ý về danh pháp: kiểu dữ liệu tổng quát mà XLA xử lý là một mảng N chiều chứa các phần tử thuộc một số kiểu thống nhất (chẳng hạn như float 32 bit). Trong toàn bộ tài liệu, mảng được sử dụng để biểu thị một mảng có chiều tùy ý. Để thuận tiện, các trường hợp đặc biệt có tên cụ thể và quen thuộc hơn; ví dụ: vectơ là mảng 1 chiều và ma trận là mảng 2 chiều.

Sau tất cả

Xem thêm XlaBuilder::AfterAll .

AfterAll lấy một số lượng mã thông báo khác nhau và tạo ra một mã thông báo duy nhất. Mã thông báo là các loại nguyên thủy có thể được xâu chuỗi giữa các hoạt động tác động phụ để thực thi đặt hàng. AfterAll có thể được sử dụng như một liên kết của các mã thông báo để đặt hàng một thao tác sau một thao tác đã đặt.

AfterAll(operands)

Tranh luận Kiểu ngữ nghĩa
operands XlaOp số lượng mã thông báo biến thiên

tất cả tập hợp

Xem thêm XlaBuilder::AllGather .

Thực hiện nối trên các bản sao.

AllGather(operand, all_gather_dim, shard_count, replica_group_ids, channel_id)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp Mảng để nối trên các bản sao.
all_gather_dim int64 Kích thước nối.
replica_groups vectơ của vectơ của int64 Các nhóm mà phép nối được thực hiện giữa chúng.
channel_id int64 tùy chọn ID kênh tùy chọn để giao tiếp giữa các mô-đun.
  • replica_groups là danh sách các nhóm bản sao mà phép nối giữa chúng được thực hiện (có thể truy xuất id bản sao cho bản sao hiện tại bằng cách sử dụng ReplicaId ). Thứ tự của các bản sao trong mỗi nhóm xác định thứ tự mà đầu vào của chúng được định vị trong kết quả. replica_groups phải trống (trong trường hợp đó, tất cả các bản sao thuộc về một nhóm duy nhất, được sắp xếp từ 0 đến N - 1 ) hoặc chứa cùng số phần tử với số lượng bản sao. Ví dụ: replica_groups = {0, 2}, {1, 3} thực hiện nối giữa các bản sao 0213 .
  • shard_count là kích thước của mỗi nhóm bản sao. Chúng tôi cần điều này trong trường hợp replica_groups trống.
  • channel_id được sử dụng để giao tiếp giữa các mô-đun: chỉ các hoạt động all-gather với cùng một channel_id mới có thể giao tiếp với nhau.

Hình dạng đầu ra là hình dạng đầu vào với all_gather_dim được làm lớn hơn shard_count lần. Ví dụ: nếu có hai bản sao và toán hạng có giá trị [1.0, 2.5][3.0, 5.25] tương ứng trên hai bản sao, thì giá trị đầu ra từ thao tác này trong đó all_gather_dim bằng 0 sẽ là [1.0, 2.5, 3.0, 5.25] trên cả hai bản sao.

Tất cảGiảm

Xem thêm XlaBuilder::AllReduce .

Thực hiện tính toán tùy chỉnh trên các bản sao.

AllReduce(operand, computation, replica_group_ids, channel_id)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp Mảng hoặc một bộ mảng không trống để rút gọn trên các bản sao.
computation XlaComputation tính toán giảm
replica_groups vectơ của vectơ của int64 Các nhóm giữa đó việc giảm được thực hiện
channel_id int64 tùy chọn ID kênh tùy chọn để liên lạc giữa các mô-đun
  • Khi operand là một bộ gồm các mảng, phép rút gọn tất cả được thực hiện trên từng phần tử của bộ.
  • replica_groups là danh sách các nhóm bản sao trong đó quá trình giảm được thực hiện (id bản sao cho bản sao hiện tại có thể được truy xuất bằng cách sử dụng ReplicaId ). replica_groups phải trống (trong trường hợp đó, tất cả các bản sao thuộc về một nhóm) hoặc chứa cùng một số phần tử với số lượng bản sao. Ví dụ: replica_groups = {0, 2}, {1, 3} thực hiện giảm giữa các bản sao 0213 .
  • channel_id được sử dụng để giao tiếp giữa các mô-đun: chỉ các hoạt động all-reduce với cùng một channel_id mới có thể giao tiếp với nhau.

Hình dạng đầu ra giống như hình dạng đầu vào. Ví dụ: nếu có hai bản sao và toán hạng có giá trị [1.0, 2.5][3.0, 5.25] tương ứng trên hai bản sao, thì giá trị đầu ra từ hoạt động này và phép tính tổng sẽ là [4.0, 7.75] trên cả hai bản sao bản sao. Nếu đầu vào là một bộ, thì đầu ra cũng là một bộ.

Việc tính toán kết quả của AllReduce yêu cầu phải có một đầu vào từ mỗi bản sao, vì vậy nếu một bản sao thực thi một nút AllReduce nhiều lần hơn một bản sao khác, thì bản sao cũ sẽ đợi mãi mãi. Vì tất cả các bản sao đều chạy cùng một chương trình nên không có nhiều cách để điều đó xảy ra, nhưng có thể xảy ra khi điều kiện của vòng lặp while phụ thuộc vào dữ liệu từ nguồn cấp dữ liệu và dữ liệu được truyền vào khiến vòng lặp lặp lại nhiều lần hơn trên một bản sao hơn một bản sao khác.

Tất cả đến tất cả

Xem thêm XlaBuilder::AllToAll .

AllToAll là một hoạt động tập thể gửi dữ liệu từ tất cả các lõi đến tất cả các lõi. Nó có hai giai đoạn:

  1. Giai đoạn tán xạ. Trên mỗi lõi, toán hạng được chia thành split_count số khối dọc theo split_dimensions và các khối được phân tán đến tất cả các lõi, ví dụ: khối thứ i được gửi đến lõi thứ i.
  2. Giai đoạn thu thập. Mỗi lõi nối các khối nhận được dọc theo concat_dimension .

Các lõi tham gia có thể được cấu hình bằng cách:

  • replica_groups : mỗi ReplicaGroup chứa danh sách id bản sao tham gia tính toán (có thể truy xuất id bản sao cho bản sao hiện tại bằng cách sử dụng ReplicaId ). AllToAll sẽ được áp dụng trong các nhóm con theo thứ tự đã chỉ định. Ví dụ: replica_groups = { {1,2,3}, {4,5,0} } có nghĩa là AllToAll sẽ được áp dụng trong các bản sao {1, 2, 3} và trong giai đoạn thu thập, đồng thời các khối nhận được sẽ được nối theo cùng thứ tự 1, 2, 3. Sau đó, một AllToAll khác sẽ được áp dụng trong các bản sao 4, 5, 0 và thứ tự nối cũng là 4, 5, 0. Nếu replica_groups trống, tất cả các bản sao thuộc về một nhóm, theo thứ tự nối của sự xuất hiện của họ.

điều kiện tiên quyết:

  • Kích thước thứ nguyên của toán hạng trên split_dimension chia hết cho split_count .
  • Hình dạng của toán hạng không phải là bộ.

AllToAll(operand, split_dimension, concat_dimension, split_count, replica_groups)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng đầu vào n chiều
split_dimension int64 Một giá trị trong khoảng [0, n) đặt tên cho thứ nguyên mà toán hạng được chia
concat_dimension int64 một giá trị trong khoảng [0, n) đặt tên cho kích thước mà các khối phân tách được nối với nhau
split_count int64 số lượng lõi tham gia hoạt động này. Nếu replica_groups trống, đây sẽ là số lượng bản sao; mặt khác, con số này phải bằng với số lượng bản sao trong mỗi nhóm.
replica_groups ReplicaGroup mỗi nhóm chứa một danh sách id bản sao.

Dưới đây là một ví dụ về Alltoall.

XlaBuilder b("alltoall");
auto x = Parameter(&b, 0, ShapeUtil::MakeShape(F32, {4, 16}), "x");
AllToAll(x, /*split_dimension=*/1, /*concat_dimension=*/0, /*split_count=*/4);

Trong ví dụ này, có 4 lõi tham gia Alltoall. Trên mỗi lõi, toán hạng được chia thành 4 phần dọc theo chiều 1, vì vậy mỗi phần có dạng f32[4,4]. 4 phần nằm rải rác cho tất cả các lõi. Sau đó, mỗi lõi nối các phần nhận được dọc theo chiều 0, theo thứ tự hoặc lõi 0-4. Vì vậy, đầu ra trên mỗi lõi có hình dạng f32[16,4].

Hàng loạtNormGrad

Xem thêm XlaBuilder::BatchNormGradtài liệu chuẩn hóa hàng loạt ban đầu để biết mô tả chi tiết về thuật toán.

Tính toán độ dốc của định mức hàng loạt.

BatchNormGrad(operand, scale, mean, variance, grad_output, epsilon, feature_index)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng n chiều cần chuẩn hóa (x)
scale XlaOp Mảng 1 chiều (\(\gamma\))
mean XlaOp Mảng 1 chiều (\(\mu\))
variance XlaOp Mảng 1 chiều (\(\sigma^2\))
grad_output XlaOp Chuyển màu được chuyển đến BatchNormTraining (\( \nabla y\))
epsilon float Giá trị Epsilon (\(\epsilon\))
feature_index int64 Chỉ mục cho thứ nguyên đặc trưng trong operand

Đối với mỗi tính năng trong thứ nguyên tính năng ( feature_index là chỉ mục cho thứ nguyên tính năng trong operand ), thao tác tính toán độ dốc đối với operand , offsetscale trên tất cả các thứ nguyên khác. feature_index phải là chỉ mục hợp lệ cho thứ nguyên đối tượng địa lý trong operand .

Ba độ dốc được xác định theo các công thức sau (giả sử mảng 4 chiều là operand và có chỉ số kích thước tính năng l , kích thước lô m và kích thước không gian wh ):

\[ \begin{split} c_l&= \frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \left( \nabla y_{ijkl} \frac{x_{ijkl} - \mu_l}{\sigma^2_l+\epsilon} \right) \\\\ d_l&= \frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \nabla y_{ijkl} \\\\ \nabla x_{ijkl} &= \frac{\gamma_{l} }{\sqrt{\sigma^2_{l}+\epsilon} } \left( \nabla y_{ijkl} - d_l - c_l (x_{ijkl} - \mu_{l}) \right) \\\\ \nabla \gamma_l &= \sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \left( \nabla y_{ijkl} \frac{x_{ijkl} - \mu_l}{\sqrt{\sigma^2_{l}+\epsilon} } \right) \\\\\ \nabla \beta_l &= \sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \nabla y_{ijkl} \end{split} \]

Giá trị meanvariance đầu vào biểu thị giá trị khoảnh khắc trên các kích thước lô và không gian.

Loại đầu ra là một bộ gồm ba tay cầm:

đầu ra Kiểu ngữ nghĩa
grad_operand XlaOp độ dốc đối với operand đầu vào (\( \nabla x\))
grad_scale XlaOp độ dốc đối với scale đầu vào (\( \nabla \gamma\))
grad_offset XlaOp độ dốc đối với offset đầu vào (\( \nabla \beta\))

Hàng loạtNormSuy luận

Xem thêm XlaBuilder::BatchNormInferencetài liệu chuẩn hóa hàng loạt ban đầu để biết mô tả chi tiết về thuật toán.

Bình thường hóa một mảng trên các kích thước lô và không gian.

BatchNormInference(operand, scale, offset, mean, variance, epsilon, feature_index)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng n chiều được chuẩn hóa
scale XlaOp mảng 1 chiều
offset XlaOp mảng 1 chiều
mean XlaOp mảng 1 chiều
variance XlaOp mảng 1 chiều
epsilon float giá trị epsilon
feature_index int64 Chỉ mục cho thứ nguyên đặc trưng trong operand

Đối với mỗi tính năng trong thứ nguyên tính năng ( feature_index là chỉ mục cho thứ nguyên tính năng trong operand ), thao tác tính toán giá trị trung bình và phương sai trên tất cả các thứ nguyên khác và sử dụng giá trị trung bình và phương sai để chuẩn hóa từng phần tử trong operand . feature_index phải là chỉ mục hợp lệ cho thứ nguyên đối tượng địa lý trong operand .

BatchNormInference tương đương với việc gọi BatchNormTraining mà không cần tính toán meanvariance cho mỗi đợt. Nó sử dụng giá meanvariance đầu vào thay vì giá trị ước tính. Mục đích của hoạt động này là để giảm độ trễ trong suy luận, do đó có tên là BatchNormInference .

Đầu ra là một mảng n chiều, được chuẩn hóa có cùng hình dạng với operand đầu vào.

ĐợtĐịnh mứcĐào tạo

Xem thêm XlaBuilder::BatchNormTrainingthe original batch normalization paper để biết mô tả chi tiết về thuật toán.

Bình thường hóa một mảng trên các kích thước lô và không gian.

BatchNormTraining(operand, scale, offset, epsilon, feature_index)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng n chiều cần chuẩn hóa (x)
scale XlaOp Mảng 1 chiều (\(\gamma\))
offset XlaOp Mảng 1 chiều (\(\beta\))
epsilon float Giá trị Epsilon (\(\epsilon\))
feature_index int64 Chỉ mục cho thứ nguyên đặc trưng trong operand

Đối với mỗi tính năng trong thứ nguyên tính năng ( feature_index là chỉ mục cho thứ nguyên tính năng trong operand ), thao tác tính toán giá trị trung bình và phương sai trên tất cả các thứ nguyên khác và sử dụng giá trị trung bình và phương sai để chuẩn hóa từng phần tử trong operand . feature_index phải là chỉ mục hợp lệ cho thứ nguyên đối tượng địa lý trong operand .

Thuật toán diễn ra như sau cho mỗi lô trong operand \(x\) chứa m phần tử với wh là kích thước của các chiều không gian (giả sử operand là một mảng 4 chiều):

  • Tính giá trị trung bình của lô \(\mu_l\) cho mỗi đối tượng l trong thứ nguyên đối tượng:\(\mu_l=\frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h x_{ijkl}\)

  • Tính phương sai lô \(\sigma^2_l\):\(\sigma^2_l=\frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h (x_{ijkl} - \mu_l)^2\)

  • Bình thường hóa, chia tỷ lệ và thay đổi:\(y_{ijkl}=\frac{\gamma_l(x_{ijkl}-\mu_l)}{\sqrt[2]{\sigma^2_l+\epsilon} }+\beta_l\)

Giá trị epsilon, thường là một số nhỏ, được thêm vào để tránh lỗi chia cho 0.

Loại đầu ra là một bộ ba XlaOp s:

đầu ra Kiểu ngữ nghĩa
output XlaOp mảng n chiều có cùng hình dạng với operand đầu vào (y)
batch_mean XlaOp mảng 1 chiều (\(\mu\))
batch_var XlaOp Mảng 1 chiều (\(\sigma^2\))

batch_meanbatch_var là những khoảnh khắc được tính trên các thứ nguyên không gian và lô bằng cách sử dụng các công thức ở trên.

BitcastChuyển đổiType

Xem thêm XlaBuilder::BitcastConvertType .

Tương tự như tf.bitcast trong TensorFlow, thực hiện thao tác bitcast theo từng phần tử từ hình dạng dữ liệu sang hình dạng mục tiêu. Kích thước đầu vào và đầu ra phải khớp nhau: ví dụ: các phần tử s32 trở thành các phần tử f32 thông qua thói quen bitcast và một phần tử s32 sẽ trở thành bốn phần tử s8 . Bitcast được thực hiện dưới dạng ép kiểu cấp thấp, vì vậy các máy có biểu diễn dấu chấm động khác nhau sẽ cho kết quả khác nhau.

BitcastConvertType(operand, new_element_type)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng loại T với độ mờ D
new_element_type PrimitiveType gõ U

Kích thước của toán hạng và hình dạng mục tiêu phải khớp với nhau, ngoại trừ kích thước cuối cùng sẽ thay đổi theo tỷ lệ của kích thước nguyên thủy trước và sau khi chuyển đổi.

Loại phần tử nguồn và đích không được là bộ.

Chuyển đổi bitcast thành loại nguyên thủy có chiều rộng khác nhau

Lệnh BitcastConvert HLO hỗ trợ trường hợp kích thước của loại phần tử đầu ra T' không bằng kích thước của phần tử đầu vào T . Vì toàn bộ hoạt động về mặt khái niệm là bitcast và không thay đổi các byte cơ bản, nên hình dạng của phần tử đầu ra phải thay đổi. Đối với B = sizeof(T), B' = sizeof(T') , có hai trường hợp có thể xảy ra.

Đầu tiên, khi B > B' , hình dạng đầu ra có kích thước nhỏ nhất mới có kích thước B/B' . Ví dụ:

  f16[10,2]{1,0} %output = f16[10,2]{1,0} bitcast-convert(f32[10]{0} %input)

Quy tắc vẫn giữ nguyên đối với vô hướng hiệu quả:

  f16[2]{0} %output = f16[2]{0} bitcast-convert(f32[] %input)

Ngoài ra, đối với B' > B , lệnh yêu cầu kích thước logic cuối cùng của hình dạng đầu vào phải bằng B'/B và kích thước này bị loại bỏ trong quá trình chuyển đổi:

  f32[10]{0} %output = f32[10]{0} bitcast-convert(f16[10,2]{1,0} %input)

Lưu ý rằng chuyển đổi giữa các băng thông bit khác nhau không theo nguyên tố.

Phát tin

Xem thêm XlaBuilder::Broadcast .

Thêm kích thước vào một mảng bằng cách sao chép dữ liệu trong mảng.

Broadcast(operand, broadcast_sizes)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp Mảng để nhân đôi
broadcast_sizes ArraySlice<int64> Kích thước của các kích thước mới

Các kích thước mới được chèn vào bên trái, tức là nếu broadcast_sizes có giá trị {a0, ..., aN} và hình dạng toán hạng có kích thước {b0, ..., bM} thì hình dạng của đầu ra có kích thước {a0, ..., aN, b0, ..., bM} .

Chỉ số thứ nguyên mới thành các bản sao của toán hạng, nghĩa là

output[i0, ..., iN, j0, ..., jM] = operand[j0, ..., jM]

Ví dụ: nếu operand là một f32 vô hướng với giá trị 2.0fbroadcast_sizes{2, 3} thì kết quả sẽ là một mảng có hình dạng f32[2, 3] và tất cả các giá trị trong kết quả sẽ là 2.0f .

Phát SóngInDim

Xem thêm XlaBuilder::BroadcastInDim .

Mở rộng kích thước và thứ hạng của một mảng bằng cách sao chép dữ liệu trong mảng.

BroadcastInDim(operand, out_dim_size, broadcast_dimensions)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp Mảng để nhân đôi
out_dim_size ArraySlice<int64> Kích thước của các kích thước của hình dạng mục tiêu
broadcast_dimensions ArraySlice<int64> Kích thước nào trong hình dạng mục tiêu, mỗi kích thước của hình dạng toán hạng tương ứng với

Tương tự như Broadcast, nhưng cho phép thêm kích thước ở bất kỳ đâu và mở rộng kích thước hiện có với kích thước 1.

operand được truyền đến hình dạng được mô tả bởi out_dim_size . broadcast_dimensions ánh xạ kích thước của operand thành kích thước của hình đích, tức là kích thước thứ i của toán hạng được ánh xạ tới kích thước thứ i của Broadcast_dimension[i] của hình dạng đầu ra. Kích thước của operand phải có kích thước 1 hoặc có cùng kích thước với kích thước trong hình dạng đầu ra mà chúng được ánh xạ tới. Các kích thước còn lại được lấp đầy với các kích thước có kích thước 1. Sau đó, phát sóng kích thước suy biến sẽ phát dọc theo các kích thước suy biến này để đạt được hình dạng đầu ra. Ngữ nghĩa được mô tả chi tiết trên trang phát sóng .

Gọi

Xem thêm XlaBuilder::Call .

Gọi một phép tính với các đối số đã cho.

Call(computation, args...)

Tranh luận Kiểu ngữ nghĩa
computation XlaComputation phép tính kiểu T_0, T_1, ..., T_{N-1} -> S với N tham số kiểu tùy ý
args dãy N XlaOp s N đối số của loại tùy ý

Arity và các loại args phải phù hợp với các tham số của computation . Nó được phép không có args .

cholesky

Xem thêm XlaBuilder::Cholesky .

Tính toán phân tách Cholesky của một loạt các ma trận xác định dương đối xứng (Hermitian).

Cholesky(a, lower)

Tranh luận Kiểu ngữ nghĩa
a XlaOp một mảng xếp hạng > 2 của một loại dấu phẩy động hoặc phức tạp.
lower bool nên sử dụng tam giác trên hay dưới của a .

Nếu lowertrue , tính toán các ma trận tam giác thấp hơn l sao cho \( a = l . l^T \). Nếu lowerfalse , tính các ma trận tam giác u sao cho \( a = u^T . u \).

Dữ liệu đầu vào chỉ được đọc từ tam giác dưới/trên của a , tùy thuộc vào giá trị của lower . Các giá trị từ tam giác khác bị bỏ qua. Dữ liệu đầu ra được trả về trong cùng một tam giác; các giá trị trong tam giác khác được xác định theo triển khai và có thể là bất kỳ giá trị nào.

Nếu hạng của a lớn hơn 2, thì a được coi là một lô ma trận, trong đó tất cả các thứ nguyên ngoại trừ 2 thứ nguyên phụ đều là thứ nguyên lô.

Nếu a không phải là xác định dương đối xứng (Hermitian), thì kết quả được xác định theo triển khai.

kẹp

Xem thêm XlaBuilder::Clamp .

Kẹp một toán hạng trong phạm vi giữa giá trị tối thiểu và tối đa.

Clamp(min, operand, max)

Tranh luận Kiểu ngữ nghĩa
min XlaOp mảng kiểu T
operand XlaOp mảng kiểu T
max XlaOp mảng kiểu T

Cho một toán hạng và các giá trị tối thiểu và tối đa, trả về toán hạng nếu nó nằm trong phạm vi giữa giá trị tối thiểu và tối đa, ngược lại trả về giá trị tối thiểu nếu toán hạng nằm dưới phạm vi này hoặc giá trị tối đa nếu toán hạng nằm trên phạm vi này. Đó là, clamp(a, x, b) = min(max(a, x), b) .

Cả ba mảng phải có cùng hình dạng. Ngoài ra, dưới dạng một hình thức phát sóng bị hạn chế, min và/hoặc max có thể là một đại lượng vô hướng thuộc loại T .

Ví dụ với vô hướng minmax :

let operand: s32[3] = {-1, 5, 9};
let min: s32 = 0;
let max: s32 = 6;
==>
Clamp(min, operand, max) = s32[3]{0, 5, 6};

Sụp đổ

Xem thêm XlaBuilder::Collapse và thao tác tf.reshape .

Thu gọn các kích thước của một mảng thành một kích thước.

Collapse(operand, dimensions)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng kiểu T
dimensions véc tơ int64 theo thứ tự, tập hợp con liên tiếp của các kích thước của T.

Thu gọn thay thế tập con đã cho của các thứ nguyên của toán hạng bằng một thứ nguyên. Các đối số đầu vào là một mảng tùy ý của loại T và một vectơ không đổi thời gian biên dịch của các chỉ số thứ nguyên. Các chỉ số thứ nguyên phải theo thứ tự (số thứ nguyên từ thấp đến cao), tập hợp con liên tiếp của thứ nguyên T. Do đó, {0, 1, 2}, {0, 1} hoặc {1, 2} đều là tập hợp thứ nguyên hợp lệ nhưng {1, 0} hoặc {0, 2} thì không. Chúng được thay thế bằng một thứ nguyên mới duy nhất, ở cùng một vị trí trong chuỗi thứ nguyên như những thứ mà chúng thay thế, với kích thước thứ nguyên mới bằng tích của các kích thước thứ nguyên ban đầu. Số thứ nguyên thấp nhất trong dimensions là thứ nguyên thay đổi chậm nhất (chính nhất) trong tổ vòng lặp làm thu gọn các thứ nguyên này và số thứ nguyên cao nhất thay đổi nhanh nhất (thứ nguyên nhỏ nhất). Xem toán tử tf.reshape nếu cần thêm thứ tự thu gọn chung.

Ví dụ, đặt v là một mảng gồm 24 phần tử:

let v = f32[4x2x3] { { {10, 11, 12},  {15, 16, 17} },
{ {20, 21, 22},  {25, 26, 27} },
{ {30, 31, 32},  {35, 36, 37} },
{ {40, 41, 42},  {45, 46, 47} } };

// Collapse to a single dimension, leaving one dimension.
let v012 = Collapse(v, {0,1,2});
then v012 == f32[24] {10, 11, 12, 15, 16, 17,
20, 21, 22, 25, 26, 27,
30, 31, 32, 35, 36, 37,
40, 41, 42, 45, 46, 47};

// Collapse the two lower dimensions, leaving two dimensions.
let v01 = Collapse(v, {0,1});
then v01 == f32[4x6] { {10, 11, 12, 15, 16, 17},
{20, 21, 22, 25, 26, 27},
{30, 31, 32, 35, 36, 37},
{40, 41, 42, 45, 46, 47} };

// Collapse the two higher dimensions, leaving two dimensions.
let v12 = Collapse(v, {1,2});
then v12 == f32[8x3] { {10, 11, 12},
{15, 16, 17},
{20, 21, 22},
{25, 26, 27},
{30, 31, 32},
{35, 36, 37},
{40, 41, 42},
{45, 46, 47} };

Tập ThểPermute

Xem thêm XlaBuilder::CollectivePermute .

CollectivePermute là một hoạt động tập thể gửi và nhận các bản sao chéo dữ liệu.

CollectivePermute(operand, source_target_pairs)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng đầu vào n chiều
source_target_pairs Vectơ <int64, int64> Danh sách các cặp (source_replica_id, target_replica_id). Đối với mỗi cặp, toán hạng được gửi từ bản sao nguồn đến bản sao đích.

Lưu ý rằng có những hạn chế sau đối với source_target_pair :

  • Hai cặp bất kỳ không được có cùng id bản sao đích và chúng không được có cùng id bản sao nguồn.
  • Nếu id bản sao không phải là mục tiêu trong bất kỳ cặp nào, thì đầu ra trên bản sao đó là một tensor bao gồm 0(s) có cùng hình dạng với đầu vào.

nối

Xem thêm XlaBuilder::ConcatInDim .

Concatenate soạn một mảng từ nhiều toán hạng mảng. Mảng có cùng thứ hạng với từng toán hạng của mảng đầu vào (phải có cùng thứ hạng với nhau) và chứa các đối số theo thứ tự mà chúng được chỉ định.

Concatenate(operands..., dimension)

Tranh luận Kiểu ngữ nghĩa
operands dãy N XlaOp N mảng kiểu T có kích thước [L0, L1,...]. Yêu cầu N >= 1.
dimension int64 Một giá trị trong khoảng [0, N) đặt tên cho thứ nguyên được nối giữa operands .

Ngoại trừ dimension tất cả các kích thước phải giống nhau. Điều này là do XLA không hỗ trợ các mảng "rách rưới". Cũng xin lưu ý rằng các giá trị xếp hạng 0 không thể được nối (vì không thể đặt tên cho thứ nguyên mà theo đó sự nối xảy ra).

ví dụ 1 chiều:

Concat({ {2, 3}, {4, 5}, {6, 7} }, 0)
>>> {2, 3, 4, 5, 6, 7}

Ví dụ 2 chiều:

let a = {
{1, 2},
{3, 4},
{5, 6},
};
let b = {
{7, 8},
};
Concat({a, b}, 0)
>>> {
{1, 2},
{3, 4},
{5, 6},
{7, 8},
}

Biểu đồ:

có điều kiện

Xem thêm XlaBuilder::Conditional .

Conditional(pred, true_operand, true_computation, false_operand, false_computation)

Tranh luận Kiểu ngữ nghĩa
pred XlaOp Vô hướng của loại PRED
true_operand XlaOp Đối số kiểu \(T_0\)
true_computation XlaComputation XlaComputation loại \(T_0 \to S\)
false_operand XlaOp Đối số kiểu \(T_1\)
false_computation XlaComputation XlaComputation của loại \(T_1 \to S\)

Thực thi true_computation nếu predtrue , false_computation nếu predfalse và trả về kết quả.

true_computation phải nhận một đối số loại \(T_0\) và sẽ được gọi với true_operand phải cùng loại. false_computation phải nhận một đối số duy nhất thuộc loại \(T_1\) và sẽ được gọi với false_operand phải cùng loại. Loại giá trị được trả về của true_computationfalse_computation phải giống nhau.

Lưu ý rằng chỉ một trong số true_computationfalse_computation sẽ được thực thi tùy thuộc vào giá trị của pred .

Conditional(branch_index, branch_computations, branch_operands)

Tranh luận Kiểu ngữ nghĩa
branch_index XlaOp Vô hướng loại S32
branch_computations dãy số N XlaComputation XlaComputings của loại \( T_0 \to S , T_1 \to S , ..., T_{N-1} \to S \)
branch_operands dãy N XlaOp Đối số kiểu \( T_0 , T_1 , ..., T_{N-1} \)

Thực thi branch_computations[branch_index] và trả về kết quả. Nếu branch_index là một S32 < 0 hoặc >= N, thì branch_computations[N-1] được thực thi như một nhánh mặc định.

Mỗi branch_computations[b] phải nhận một đối số loại T_b và sẽ được gọi với branch_operands[b] phải cùng loại. Loại giá trị trả về của mỗi branch_computations[b] phải giống nhau.

Lưu ý rằng chỉ một trong số branch_computations sẽ được thực thi tùy thuộc vào giá trị của branch_index .

Chuyển đổi (tích chập)

Xem thêm XlaBuilder::Conv .

Là ConvWithGeneralPadding, nhưng phần đệm được chỉ định theo cách ngắn gọn là CÙNG hoặc HỢP LỆ. Phần đệm CÙNG đệm đầu vào ( lhs ) bằng các số 0 để đầu ra có hình dạng giống như đầu vào khi không tính đến sải chân. Phần đệm HỢP LỆ đơn giản có nghĩa là không có phần đệm.

ConvWithGeneralPadding (tích chập)

Xem thêm XlaBuilder::ConvWithGeneralPadding .

Tính toán một tích chập thuộc loại được sử dụng trong mạng thần kinh. Ở đây, một tích chập có thể được coi là một cửa sổ n chiều di chuyển qua một vùng cơ sở n chiều và một phép tính được thực hiện cho từng vị trí có thể có của cửa sổ.

Tranh luận Kiểu ngữ nghĩa
lhs XlaOp xếp hạng n+2 mảng đầu vào
rhs XlaOp xếp hạng n+2 mảng trọng số hạt nhân
window_strides ArraySlice<int64> mảng nd của kernel
padding ArraySlice< pair<int64, int64>> mảng đệm thứ (thấp, cao)
lhs_dilation ArraySlice<int64> mảng hệ số giãn nd lhs
rhs_dilation ArraySlice<int64> mảng hệ số giãn nd rhs
feature_group_count int64 Số lượng nhóm tính năng
batch_group_count int64 số nhóm lô

Gọi n là số chiều không gian. Đối số lhs là một mảng hạng n+2 mô tả diện tích cơ sở. Đây được gọi là đầu vào, mặc dù tất nhiên rhs cũng là một đầu vào. Trong một mạng thần kinh, đây là những kích hoạt đầu vào. Kích thước n+2 là, theo thứ tự này:

  • batch : Mỗi tọa độ trong thứ nguyên này đại diện cho một đầu vào độc lập mà phép tích chập được thực hiện.
  • z/depth/features : Mỗi vị trí (y,x) trong vùng cơ sở có một vectơ liên kết với nó, đi vào chiều này.
  • spatial_dims : Mô tả n kích thước không gian xác định khu vực cơ sở mà cửa sổ di chuyển qua.

Đối số rhs là một mảng hạng n+2 mô tả bộ lọc tích chập/hạt nhân/cửa sổ. Kích thước là, theo thứ tự này:

  • output-z : Kích thước z của đầu ra.
  • input-z : Kích thước của thứ nguyên này nhân feature_group_count phải bằng kích thước của thứ nguyên z tính bằng lhs.
  • spatial_dims : Mô tả n kích thước không gian xác định cửa sổ thứ di chuyển qua khu vực cơ sở.

Đối số window_strides chỉ định bước tiến của cửa sổ tích chập trong các chiều không gian. Ví dụ: nếu sải chân trong chiều không gian đầu tiên là 3, thì cửa sổ chỉ có thể được đặt tại các tọa độ mà chỉ số không gian đầu tiên chia hết cho 3.

Đối số padding chỉ định số lượng đệm bằng 0 được áp dụng cho vùng cơ sở. Số lượng phần đệm có thể âm -- giá trị tuyệt đối của phần đệm âm cho biết số phần tử cần xóa khỏi thứ nguyên đã chỉ định trước khi thực hiện tích chập. padding[0] chỉ định phần đệm cho thứ nguyên ypadding[1] chỉ định phần đệm cho thứ nguyên x . Mỗi cặp có phần đệm thấp là phần tử đầu tiên và phần đệm cao là phần tử thứ hai. Phần đệm thấp được áp dụng theo hướng của các chỉ số thấp hơn trong khi phần đệm cao được áp dụng theo hướng của các chỉ số cao hơn. Ví dụ: nếu padding[1](2,3) thì sẽ có phần đệm bằng 2 số 0 ở bên trái và 3 số 0 ở bên phải trong chiều không gian thứ hai. Sử dụng phần đệm tương đương với việc chèn các giá trị 0 giống nhau đó vào đầu vào ( lhs ) trước khi thực hiện tích chập.

Các đối số lhs_dilationrhs_dilation chỉ định hệ số giãn nở được áp dụng cho lhs và rhs tương ứng trong mỗi chiều không gian. Nếu hệ số giãn nở trong một chiều không gian là d, thì các lỗ d-1 được đặt hoàn toàn giữa mỗi mục trong chiều đó, làm tăng kích thước của mảng. Các lỗ được lấp đầy bằng một giá trị không hoạt động, mà đối với tích chập có nghĩa là số không.

Sự giãn nở của rhs còn được gọi là tích chập atrous. Để biết thêm chi tiết, hãy xem tf.nn.atrous_conv2d . Sự giãn nở của lhs còn được gọi là tích chập chuyển vị. Để biết thêm chi tiết, hãy xem tf.nn.conv2d_transpose .

Đối số feature_group_count (giá trị mặc định 1) có thể được sử dụng cho các tích chập được nhóm. feature_group_count cần phải là ước số của cả thứ nguyên đối tượng đầu vào và đầu ra. Nếu feature_group_count lớn hơn 1, điều đó có nghĩa là về mặt khái niệm, thứ nguyên đối tượng đầu vào và đầu ra cũng như thứ nguyên đối tượng đầu ra rhs được chia đều thành feature_group_count nhiều nhóm, mỗi nhóm bao gồm một chuỗi các đối tượng liên tiếp. Thứ nguyên tính năng đầu vào của rhs cần phải bằng thứ nguyên tính năng đầu vào lhs chia cho feature_group_count (để nó có kích thước của một nhóm tính năng đầu vào). Các nhóm thứ i được sử dụng cùng nhau để tính toán feature_group_count nhiều tích chập riêng biệt. Kết quả của các tích chập này được nối với nhau trong thứ nguyên tính năng đầu ra.

Đối với tích chập theo chiều sâu, đối số feature_group_count sẽ được đặt thành thứ nguyên tính năng đầu vào và bộ lọc sẽ được định hình lại từ [filter_height, filter_width, in_channels, channel_multiplier] thành [filter_height, filter_width, 1, in_channels * channel_multiplier] . Để biết thêm chi tiết, hãy xem tf.nn.depthwise_conv2d .

Đối số batch_group_count (giá trị mặc định 1) có thể được sử dụng cho các bộ lọc được nhóm trong quá trình lan truyền ngược. batch_group_count cần phải là ước số của kích thước của thứ nguyên lô lhs (đầu vào). Nếu batch_group_count lớn hơn 1, điều đó có nghĩa là kích thước lô đầu ra phải bằng kích thước input batch / batch_group_count . batch_group_count phải là ước số của kích thước tính năng đầu ra.

Hình dạng đầu ra có các kích thước này, theo thứ tự sau:

  • batch : Kích thước của thứ nguyên này nhân với batch_group_count phải bằng kích thước của thứ nguyên batch bằng lhs.
  • z : Cùng kích thước với output-z trên kernel ( rhs ).
  • spatial_dims : Một giá trị cho mỗi vị trí hợp lệ của cửa sổ tích chập.

Hình trên cho thấy cách hoạt động của trường batch_group_count . Thực tế, chúng tôi chia từng lô lhs thành các nhóm batch_group_count và thực hiện tương tự đối với các tính năng đầu ra. Sau đó, đối với mỗi nhóm này, chúng tôi thực hiện tích chập theo cặp và nối đầu ra dọc theo thứ nguyên tính năng đầu ra. Ngữ nghĩa hoạt động của tất cả các kích thước khác (tính năng và không gian) vẫn giữ nguyên.

Các vị trí hợp lệ của cửa sổ tích chập được xác định bởi các bước và kích thước của vùng cơ sở sau khi đệm.

Để mô tả chức năng của một tích chập, hãy xem xét một tích chập 2d và chọn một số batch , z , y , x cố định trong đầu ra. Khi đó (y,x) là vị trí của một góc cửa sổ trong vùng cơ sở (ví dụ: góc trên bên trái, tùy thuộc vào cách bạn diễn giải các kích thước không gian). Bây giờ chúng ta có một cửa sổ 2d, được lấy từ vùng cơ sở, trong đó mỗi điểm 2d được liên kết với một vectơ 1d, vì vậy chúng ta có một hộp 3d. Từ nhân tích chập, vì chúng ta đã cố định tọa độ đầu ra z , nên chúng ta cũng có một hộp 3d. Hai hộp có cùng kích thước, vì vậy chúng ta có thể lấy tổng các tích phần tử giữa hai hộp (tương tự như tích vô hướng). Đó là giá trị đầu ra.

Lưu ý rằng nếu output-z là ví dụ: 5, thì mỗi vị trí của cửa sổ sẽ tạo ra 5 giá trị trong đầu ra thành thứ nguyên z của đầu ra. Các giá trị này khác nhau ở phần nào của hạt nhân tích chập được sử dụng - có một hộp giá trị 3d riêng biệt được sử dụng cho mỗi tọa độ output-z . Vì vậy, bạn có thể coi nó là 5 kết cấu riêng biệt với một bộ lọc khác nhau cho mỗi kết cấu.

Đây là mã giả cho tích chập 2d với đệm và sải chân:

for (b, oz, oy, ox) {  // output coordinates
  value = 0;
  for (iz, ky, kx) {  // kernel coordinates and input z
    iy = oy*stride_y + ky - pad_low_y;
    ix = ox*stride_x + kx - pad_low_x;
    if ((iy, ix) inside the base area considered without padding) {
      value += input(b, iz, iy, ix) * kernel(oz, iz, ky, kx);
    }
  }
  output(b, oz, oy, ox) = value;
}

Chuyển đổiElementType

Xem thêm XlaBuilder::ConvertElementType .

Tương tự như static_cast theo từng phần tử trong C++, thực hiện thao tác chuyển đổi theo từng phần tử từ hình dạng dữ liệu sang hình dạng mục tiêu. Kích thước phải phù hợp và chuyển đổi là yếu tố khôn ngoan; ví dụ: phần tử s32 trở thành phần tử f32 thông qua quy trình chuyển đổi s32 -to- f32 .

ConvertElementType(operand, new_element_type)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp mảng loại T với độ mờ D
new_element_type PrimitiveType gõ U

Kích thước của toán hạng và hình dạng mục tiêu phải phù hợp. Loại phần tử nguồn và đích không được là bộ.

Một chuyển đổi chẳng hạn như T=s32 thành U=f32 sẽ thực hiện quy trình chuyển đổi int-to-float chuẩn hóa, chẳng hạn như làm tròn thành chẵn.

let a: s32[3] = {0, 1, 2};
let b: f32[3] = convert(a, f32);
then b == f32[3]{0.0, 1.0, 2.0}

QuaBản SaoTổng

Thực hiện AllReduce với phép tính tổng.

Cuộc gọi tùy chỉnh

Xem thêm XlaBuilder::CustomCall .

Gọi một hàm do người dùng cung cấp trong một tính toán.

CustomCall(target_name, args..., shape)

Tranh luận Kiểu ngữ nghĩa
target_name string Tên của chức năng. Một lệnh gọi sẽ được phát ra nhắm vào tên biểu tượng này.
args dãy N XlaOp s N đối số thuộc loại tùy ý, sẽ được truyền cho hàm.
shape Shape Hình dạng đầu ra của hàm

Chữ ký hàm là như nhau, bất kể arity hoặc loại đối số:

extern "C" void target_name(void* out, void** in);

Ví dụ: nếu CustomCall được sử dụng như sau:

let x = f32[2] {1,2};
let y = f32[2x3] { {10, 20, 30}, {40, 50, 60} };

CustomCall("myfunc", {x, y}, f32[3x3])

Đây là một ví dụ về việc triển khai myfunc :

extern "C" void myfunc(void* out, void** in) {
  float (&x)[2] = *static_cast<float(*)[2]>(in[0]);
  float (&y)[2][3] = *static_cast<float(*)[2][3]>(in[1]);
  EXPECT_EQ(1, x[0]);
  EXPECT_EQ(2, x[1]);
  EXPECT_EQ(10, y[0][0]);
  EXPECT_EQ(20, y[0][1]);
  EXPECT_EQ(30, y[0][2]);
  EXPECT_EQ(40, y[1][0]);
  EXPECT_EQ(50, y[1][1]);
  EXPECT_EQ(60, y[1][2]);
  float (&z)[3][3] = *static_cast<float(*)[3][3]>(out);
  z[0][0] = x[1] + y[1][0];
  // ...
}

Chức năng do người dùng cung cấp không được có tác dụng phụ và việc thực thi của nó phải bình thường.

chấm

Xem thêm XlaBuilder::Dot .

Dot(lhs, rhs)

Tranh luận Kiểu ngữ nghĩa
lhs XlaOp mảng kiểu T
rhs XlaOp mảng kiểu T

Ngữ nghĩa chính xác của thao tác này phụ thuộc vào cấp bậc của các toán hạng:

Đầu vào đầu ra ngữ nghĩa
vectơ [n] dot chấm vectơ [n] vô hướng sản phẩm chấm vector
ma trận [mxk] dot chấm [k] véc tơ [m] phép nhân ma trận-vector
ma trận [mxk] ma trận dot [kxn] ma trận [mxn] phép nhân ma trận

Hoạt động thực hiện tổng các tích trên chiều thứ hai của lhs (hoặc thứ nhất nếu nó có thứ hạng 1) và chiều thứ nhất của rhs . Đây là các kích thước "được ký hợp đồng". Các kích thước được ký hợp đồng của lhsrhs phải có cùng kích thước. Trong thực tế, nó có thể được sử dụng để thực hiện tích giữa các vectơ, phép nhân vectơ/ma trận hoặc phép nhân ma trận/ma trận.

chấm chung

Xem thêm XlaBuilder::DotGeneral .

DotGeneral(lhs, rhs, dimension_numbers)

Tranh luận Kiểu ngữ nghĩa
lhs XlaOp mảng kiểu T
rhs XlaOp mảng kiểu T
dimension_numbers DotDimensionNumbers hợp đồng và số thứ nguyên lô

Là Dấu chấm, nhưng cho phép chỉ định số thứ nguyên lô và hợp đồng cho cả 'lhs' và 'rhs'.

Trường Kích thước Dấu chấm Kiểu ngữ nghĩa
'lhs_contracting_dimensions' lặp lại int64 số thứ nguyên hợp đồng 'lhs'
'rhs_contracting_dimensions' lặp lại int64 số thứ nguyên hợp đồng 'rhs'
'lhs_batch_dimensions' lặp lại int64 số thứ nguyên lô 'lhs'
'rhs_batch_dimensions' lặp lại int64 số thứ nguyên lô 'rhs'

DotGeneral thực hiện tổng sản phẩm trên các tham số hợp đồng được chỉ định trong 'dimension_numbers'.

Các số thứ nguyên hợp đồng được liên kết từ 'lhs' và 'rhs' không cần phải giống nhau nhưng phải có cùng kích thước thứ nguyên.

Ví dụ với số thứ nguyên hợp đồng:

lhs = { {1.0, 2.0, 3.0},
{4.0, 5.0, 6.0} }

rhs = { {1.0, 1.0, 1.0},
{2.0, 2.0, 2.0} }

DotDimensionNumbers dnums;
dnums.add_lhs_contracting_dimensions(1);
dnums.add_rhs_contracting_dimensions(1);

DotGeneral(lhs, rhs, dnums) -> { {6.0, 12.0},
{15.0, 30.0} }

Số thứ nguyên lô được liên kết từ 'lhs' và 'rhs' phải có cùng kích thước thứ nguyên.

Ví dụ với số thứ nguyên lô (kích thước lô 2, ma trận 2x2):

lhs = { { {1.0, 2.0},
{3.0, 4.0} },
{ {5.0, 6.0},
{7.0, 8.0} } }

rhs = { { {1.0, 0.0},
{0.0, 1.0} },
{ {1.0, 0.0},
{0.0, 1.0} } }

DotDimensionNumbers dnums;
dnums.add_lhs_contracting_dimensions(2);
dnums.add_rhs_contracting_dimensions(1);
dnums.add_lhs_batch_dimensions(0);
dnums.add_rhs_batch_dimensions(0);

DotGeneral(lhs, rhs, dnums) -> { { {1.0, 2.0},
{3.0, 4.0} },
{ {5.0, 6.0},
{7.0, 8.0} } }
Đầu vào đầu ra ngữ nghĩa
[b0, m, k] dot [b0, k, n] [b0,m,n] lô matmul
[b0, b1, m, k] dot [b0, b1, k, n] [b0, b1, m, n] lô matmul

Theo đó, số thứ nguyên kết quả bắt đầu bằng thứ nguyên lô, sau đó là thứ nguyên không hợp đồng/không theo lô 'lhs' và cuối cùng là thứ nguyên không hợp đồng/không theo lô 'rhs'.

DynamicSlice

Xem thêm XlaBuilder::DynamicSlice .

DynamicSlice trích xuất một mảng phụ từ mảng đầu vào tại start_indices động. Kích thước của lát cắt trong mỗi thứ nguyên được chuyển vào size_indices , xác định điểm kết thúc của các khoảng thời gian lát cắt riêng trong mỗi thứ nguyên: [bắt đầu, bắt đầu + kích thước). Hình dạng của start_indices phải có thứ hạng == 1, với kích thước thứ nguyên bằng thứ hạng của operand .

DynamicSlice(operand, start_indices, size_indices)

Tranh luận Kiểu ngữ nghĩa
operand XlaOp Mảng N chiều kiểu T
start_indices dãy N XlaOp Danh sách N số nguyên vô hướng chứa các chỉ số bắt đầu của lát cho mỗi thứ nguyên. Value must be greater than or equal to zero.
size_indices ArraySlice<int64> List of N integers containing the slice size for each dimension. Each value must be strictly greater than zero, and start + size must be less than or equal to the size of the dimension to avoid wrapping modulo dimension size.

The effective slice indices are computed by applying the following transformation for each index i in [1, N) before performing the slice:

start_indices[i] = clamp(start_indices[i], 0, operand.dimension_size[i] - size_indices[i])

This ensures that the extracted slice is always in-bounds with respect to the operand array. If the slice is in-bounds before the transformation is applied, the transformation has no effect.

1-dimensional example:

let a = {0.0, 1.0, 2.0, 3.0, 4.0}
let s = {2}

DynamicSlice(a, s, {2}) produces:
{2.0, 3.0}

2-dimensional example:

let b =
{ {0.0,  1.0,  2.0},
{3.0,  4.0,  5.0},
{6.0,  7.0,  8.0},
{9.0, 10.0, 11.0} }
let s = {2, 1}

DynamicSlice(b, s, {2, 2}) produces:
{ { 7.0,  8.0},
{10.0, 11.0} }

DynamicUpdateSlice

See also XlaBuilder::DynamicUpdateSlice .

DynamicUpdateSlice generates a result which is the value of the input array operand , with a slice update overwritten at start_indices . The shape of update determines the shape of the sub-array of the result which is updated. The shape of start_indices must be rank == 1, with dimension size equal to the rank of operand .

DynamicUpdateSlice(operand, update, start_indices)

Arguments Type Semantics
operand XlaOp N dimensional array of type T
update XlaOp N dimensional array of type T containing the slice update. Each dimension of update shape must be strictly greater than zero, and start + update must be less than or equal to the operand size for each dimension to avoid generating out-of-bounds update indices.
start_indices sequence of N XlaOp List of N scalar integers containing the starting indices of the slice for each dimension. Value must be greater than or equal to zero.

The effective slice indices are computed by applying the following transformation for each index i in [1, N) before performing the slice:

start_indices[i] = clamp(start_indices[i], 0, operand.dimension_size[i] - update.dimension_size[i])

This ensures that the updated slice is always in-bounds with respect to the operand array. If the slice is in-bounds before the transformation is applied, the transformation has no effect.

1-dimensional example:

let a = {0.0, 1.0, 2.0, 3.0, 4.0}
let u = {5.0, 6.0}
let s = {2}

DynamicUpdateSlice(a, u, s) produces:
{0.0, 1.0, 5.0, 6.0, 4.0}

2-dimensional example:

let b =
{ {0.0,  1.0,  2.0},
{3.0,  4.0,  5.0},
{6.0,  7.0,  8.0},
{9.0, 10.0, 11.0} }
let u =
{ {12.0,  13.0},
{14.0,  15.0},
{16.0,  17.0} }

let s = {1, 1}

DynamicUpdateSlice(b, u, s) produces:
{ {0.0,  1.0,  2.0},
{3.0, 12.0, 13.0},
{6.0, 14.0, 15.0},
{9.0, 16.0, 17.0} }

Element-wise binary arithmetic operations

See also XlaBuilder::Add .

A set of element-wise binary arithmetic operations is supported.

Op(lhs, rhs)

Where Op is one of Add (addition), Sub (subtraction), Mul (multiplication), Div (division), Rem (remainder), Max (maximum), Min (minimum), LogicalAnd (logical AND), or LogicalOr (logical OR).

Arguments Type Semantics
lhs XlaOp left-hand-side operand: array of type T
rhs XlaOp right-hand-side operand: array of type T

The arguments' shapes have to be either similar or compatible. See the broadcasting documentation about what it means for shapes to be compatible. The result of an operation has a shape which is the result of broadcasting the two input arrays. In this variant, operations between arrays of different ranks are not supported, unless one of the operands is a scalar.

When Op is Rem , the sign of the result is taken from the dividend, and the absolute value of the result is always less than the divisor's absolute value.

Integer division overflow (signed/unsigned division/remainder by zero or signed division/remainder of INT_SMIN with -1 ) produces an implementation defined value.

An alternative variant with different-rank broadcasting support exists for these operations:

Op(lhs, rhs, broadcast_dimensions)

Where Op is the same as above. This variant of the operation should be used for arithmetic operations between arrays of different ranks (such as adding a matrix to a vector).

The additional broadcast_dimensions operand is a slice of integers used to expand the rank of the lower-rank operand up to the rank of the higher-rank operand. broadcast_dimensions maps the dimensions of the lower-rank shape to the dimensions of the higher-rank shape. The unmapped dimensions of the expanded shape are filled with dimensions of size one. Degenerate-dimension broadcasting then broadcasts the shapes along these degenerate dimensions to equalize the shapes of both operands. The semantics are described in detail on the broadcasting page .

Element-wise comparison operations

See also XlaBuilder::Eq .

A set of standard element-wise binary comparison operations is supported. Note that standard IEEE 754 floating-point comparison semantics apply when comparing floating-point types.

Op(lhs, rhs)

Where Op is one of Eq (equal-to), Ne (not equal-to), Ge (greater-or-equal-than), Gt (greater-than), Le (less-or-equal-than), Lt (less-than). Another set of operators, EqTotalOrder, NeTotalOrder, GeTotalOrder, GtTotalOrder, LeTotalOrder, and LtTotalOrder, provide the same functionalities, except that they additionally support a total order over the floating point numbers, by enforcing -NaN < -Inf < -Finite < -0 < +0 < +Finite < +Inf < +NaN.

Arguments Type Semantics
lhs XlaOp left-hand-side operand: array of type T
rhs XlaOp right-hand-side operand: array of type T

The arguments' shapes have to be either similar or compatible. See the broadcasting documentation about what it means for shapes to be compatible. The result of an operation has a shape which is the result of broadcasting the two input arrays with the element type PRED . In this variant, operations between arrays of different ranks are not supported, unless one of the operands is a scalar.

An alternative variant with different-rank broadcasting support exists for these operations:

Op(lhs, rhs, broadcast_dimensions)

Where Op is the same as above. This variant of the operation should be used for comparison operations between arrays of different ranks (such as adding a matrix to a vector).

The additional broadcast_dimensions operand is a slice of integers specifying the dimensions to use for broadcasting the operands. The semantics are described in detail on the broadcasting page .

Element-wise unary functions

XlaBuilder supports these element-wise unary functions:

Abs(operand) Element-wise abs x -> |x| .

Ceil(operand) Element-wise ceil x -> ⌈x⌉ .

Cos(operand) Element-wise cosine x -> cos(x) .

Exp(operand) Element-wise natural exponential x -> e^x .

Floor(operand) Element-wise floor x -> ⌊x⌋ .

Imag(operand) Element-wise imaginary part of a complex (or real) shape. x -> imag(x) . If the operand is a floating point type, returns 0.

IsFinite(operand) Tests whether each element of operand is finite, ie, is not positive or negative infinity, and is not NaN . Returns an array of PRED values with the same shape as the input, where each element is true if and only if the corresponding input element is finite.

Log(operand) Element-wise natural logarithm x -> ln(x) .

LogicalNot(operand) Element-wise logical not x -> !(x) .

Logistic(operand) Element-wise logistic function computation x -> logistic(x) .

PopulationCount(operand) Computes the number of bits set in each element of operand .

Neg(operand) Element-wise negation x -> -x .

Real(operand) Element-wise real part of a complex (or real) shape. x -> real(x) . If the operand is a floating point type, returns the same value.

Rsqrt(operand) Element-wise reciprocal of square root operation x -> 1.0 / sqrt(x) .

Sign(operand) Element-wise sign operation x -> sgn(x) where

\[\text{sgn}(x) = \begin{cases} -1 & x < 0\\ -0 & x = -0\\ NaN & x = NaN\\ +0 & x = +0\\ 1 & x > 0 \end{cases}\]

using the comparison operator of the element type of operand .

Sqrt(operand) Element-wise square root operation x -> sqrt(x) .

Cbrt(operand) Element-wise cubic root operation x -> cbrt(x) .

Tan(operand) Element-wise tangent x -> tan(x) .

Tanh(operand) Element-wise hyperbolic tangent x -> tanh(x) .

Round(operand) Element-wise rounding, ties away from zero.

RoundNearestEven(operand) Element-wise rounding, ties to nearest even.

Arguments Type Semantics
operand XlaOp The operand to the function

The function is applied to each element in the operand array, resulting in an array with the same shape. It is allowed for operand to be a scalar (rank 0).

Fft

The XLA FFT operation implements the forward and inverse Fourier Transforms for real and complex inputs/outputs. Multidimensional FFTs on up to 3 axes are supported.

See also XlaBuilder::Fft .

Arguments Type Semantics
operand XlaOp The array we are Fourier transforming.
fft_type FftType See the table below.
fft_length ArraySlice<int64> The time-domain lengths of the axes being transformed. This is needed in particular for IRFFT to right-size the innermost axis, since RFFT(fft_length=[16]) has the same output shape as RFFT(fft_length=[17]) .
FftType Semantics
FFT Forward complex-to-complex FFT. Shape is unchanged.
IFFT Inverse complex-to-complex FFT. Shape is unchanged.
RFFT Forward real-to-complex FFT. Shape of the innermost axis is reduced to fft_length[-1] // 2 + 1 if fft_length[-1] is a non-zero value, omitting the reversed conjugate part of the transformed signal beyond the Nyquist frequency.
IRFFT Inverse real-to-complex FFT (ie takes complex, returns real). Shape of the innermost axis is expanded to fft_length[-1] if fft_length[-1] is a non-zero value, inferring the part of the transformed signal beyond the Nyquist frequency from the reverse conjugate of the 1 to fft_length[-1] // 2 + 1 entries.

Multidimensional FFT

When more than 1 fft_length is provided, this is equivalent to applying a cascade of FFT operations to each of the innermost axes. Note that for the real->complex and complex->real cases, the innermost axis transform is (effectively) performed first (RFFT; last for IRFFT), which is why the innermost axis is the one which changes size. Other axis transforms will then be complex->complex.

Implementation details

CPU FFT is backed by Eigen's TensorFFT. GPU FFT uses cuFFT.

Gather

The XLA gather operation stitches together several slices (each slice at a potentially different runtime offset) of an input array.

General Semantics

See also XlaBuilder::Gather . For a more intuitive description, see the "Informal Description" section below.

gather(operand, start_indices, offset_dims, collapsed_slice_dims, slice_sizes, start_index_map)

Arguments Type Semantics
operand XlaOp The array we're gathering from.
start_indices XlaOp Array containing the starting indices of the slices we gather.
index_vector_dim int64 The dimension in start_indices that "contains" the starting indices. See below for a detailed description.
offset_dims ArraySlice<int64> The set of dimensions in the output shape that offset into an array sliced from operand.
slice_sizes ArraySlice<int64> slice_sizes[i] is the bounds for the slice on dimension i .
collapsed_slice_dims ArraySlice<int64> The set of dimensions in each slice that are collapsed away. These dimensions must have size 1.
start_index_map ArraySlice<int64> A map that describes how to map indices in start_indices to legal indices into operand.
indices_are_sorted bool Whether the indices are guaranteed to be sorted by the caller.
unique_indices bool Whether the indices are guaranteed to be unique by the caller.

For convenience, we label dimensions in the output array not in offset_dims as batch_dims .

The output is an array of rank batch_dims.size + offset_dims.size .

The operand.rank must equal the sum of offset_dims.size and collapsed_slice_dims.size . Also, slice_sizes.size has to be equal to operand.rank .

If index_vector_dim is equal to start_indices.rank we implicitly consider start_indices to have a trailing 1 dimension (ie if start_indices was of shape [6,7] and index_vector_dim is 2 then we implicitly consider the shape of start_indices to be [6,7,1] ).

The bounds for the output array along dimension i is computed as follows:

  1. If i is present in batch_dims (ie is equal to batch_dims[k] for some k ) then we pick the corresponding dimension bounds out of start_indices.shape , skipping index_vector_dim (ie pick start_indices.shape.dims [ k ] if k < index_vector_dim and start_indices.shape.dims [ k + 1 ] otherwise).

  2. If i is present in offset_dims (ie equal to offset_dims [ k ] for some k ) then we pick the corresponding bound out of slice_sizes after accounting for collapsed_slice_dims (ie we pick adjusted_slice_sizes [ k ] where adjusted_slice_sizes is slice_sizes with the bounds at indices collapsed_slice_dims removed).

Formally, the operand index In corresponding to a given output index Out is calculated as follows:

  1. Let G = { Out [ k ] for k in batch_dims }. Use G to slice out a vector S such that S [ i ] = start_indices [Combine( G , i )] where Combine(A, b) inserts b at position index_vector_dim into A. Note that this is well defined even if G is empty -- if G is empty then S = start_indices .

  2. Create a starting index, S in , into operand using S by scattering S using start_index_map . More precisely:

    1. S in [ start_index_map [ k ]] = S [ k ] if k < start_index_map.size .

    2. S in [ _ ] = 0 otherwise.

  3. Create an index O in into operand by scattering the indices at the offset dimensions in Out according to the collapsed_slice_dims set. More precisely:

    1. O in [ remapped_offset_dims ( k )] = Out [ offset_dims [ k ]] if k < offset_dims.size ( remapped_offset_dims is defined below).

    2. O in [ _ ] = 0 otherwise.

  4. In is O in + S in where + is element-wise addition.

remapped_offset_dims is a monotonic function with domain [ 0 , offset_dims.size ) and range [ 0 , operand.rank ) \ collapsed_slice_dims . So if, eg, offset_dims.size is 4 , operand.rank is 6 and collapsed_slice_dims is { 0 , 2 } then remapped_offset_dims is { 01 , 13 , 24 , 35 }.

If indices_are_sorted is set to true then XLA can assume that start_indices are sorted (in ascending start_index_map order) by the user. If they are not then the semantics is implementation defined.

If unique_indices is set to true then XLA can assume that all element scattered to are unique. So XLA could use non-atomic operations. If unique_indices is set to true and the indices being scattered to are not unique then the semantics is implementation defined.

Informal Description and Examples

Informally, every index Out in the output array corresponds to an element E in the operand array, computed as follows:

  • We use the batch dimensions in Out to look up a starting index from start_indices .

  • We use start_index_map to map the starting index (whose size may be less than operand.rank) to a "full" starting index into the operand .

  • We dynamic-slice out a slice with size slice_sizes using the full starting index.

  • We reshape the slice by collapsing the collapsed_slice_dims dimensions. Since all collapsed slice dimensions must have a bound of 1, this reshape is always legal.

  • We use the offset dimensions in Out to index into this slice to get the input element, E , corresponding to output index Out .

index_vector_dim is set to start_indices.rank - 1 in all of the examples that follow. More interesting values for index_vector_dim do not change the operation fundamentally, but make the visual representation more cumbersome.

To get an intuition on how all of the above fits together, let's look at an example that gathers 5 slices of shape [8,6] from a [16,11] array. The position of a slice into the [16,11] array can be represented as an index vector of shape S64[2] , so the set of 5 positions can be represented as a S64[5,2] array.

The behavior of the gather operation can then be depicted as an index transformation that takes [ G , O 0 , O 1 ], an index in the output shape, and maps it to an element in the input array in the following way:

We first select an ( X , Y ) vector from the gather indices array using G . The element in the output array at index [ G , O 0 , O 1 ] is then the element in the input array at index [ X + O 0 , Y + O 1 ].

slice_sizes is [8,6] , which decides the range of O 0 and O 1 , and this in turn decides the bounds of the slice.

This gather operation acts as a batch dynamic slice with G as the batch dimension.

The gather indices may be multidimensional. For instance, a more general version of the example above using a "gather indices" array of shape [4,5,2] would translate indices like this:

Again, this acts as a batch dynamic slice G 0 and G 1 as the batch dimensions. The slice size is still [8,6] .

The gather operation in XLA generalizes the informal semantics outlined above in the following ways:

  1. We can configure which dimensions in the output shape are the offset dimensions (dimensions containing O 0 , O 1 in the last example). The output batch dimensions (dimensions containing G 0 , G 1 in the last example) are defined to be the output dimensions that are not offset dimensions.

  2. The number of output offset dimensions explicitly present in the output shape may be smaller than the input rank. These "missing" dimensions, which are listed explicitly as collapsed_slice_dims , must have a slice size of 1 . Since they have a slice size of 1 the only valid index for them is 0 and eliding them does not introduce ambiguity.

  3. The slice extracted from the "Gather Indices" array (( X , Y ) in the last example) may have fewer elements than the input array rank, and an explicit mapping dictates how the index should be expanded to have the same rank as the input.

As a final example, we use (2) and (3) to implement tf.gather_nd :

G 0 and G 1 are used to slice out a starting index from the gather indices array as usual, except the starting index has only one element, X . Similarly, there is only one output offset index with the value O 0 . However, before being used as indices into the input array, these are expanded in accordance to "Gather Index Mapping" ( start_index_map in the formal description) and "Offset Mapping" ( remapped_offset_dims in the formal description) into [ X , 0 ] and [ 0 , O 0 ] respectively, adding up to [ X , O 0 ]. In other words, the output index [ G 0 , G 1 , O 0 ] maps to the input index [ GatherIndices [ G 0 , G 1 , 0 ], O 0 ] which gives us the semantics for tf.gather_nd .

slice_sizes for this case is [1,11] . Intuitively this means that every index X in the gather indices array picks an entire row and the result is the concatenation of all these rows.

GetDimensionSize

See also XlaBuilder::GetDimensionSize .

Returns the size of the given dimension of the operand. The operand must be array shaped.

GetDimensionSize(operand, dimension)

Arguments Type Semantics
operand XlaOp n dimensional input array
dimension int64 A value in the interval [0, n) that specifies the dimension

SetDimensionSize

See also XlaBuilder::SetDimensionSize .

Sets the dynamic size of XlaOp's given dimension. The operand must be array shaped.

SetDimensionSize(operand, size, dimension)

Arguments Type Semantics
operand XlaOp n dimensional input array.
size XlaOp int32 representing the runtime dynamic size.
dimension int64 A value in the interval [0, n) that specifies the dimension.

Pass through the operand as result, with dynamic dimension tracked by the compiler.

Padded values will be ignored by downstream reduction ops.

let v: f32[10] = f32[10]{1, 2, 3, 4, 5, 6, 7, 8, 9, 10};
let five: s32 = 5;
let six: s32 = 6;

// Setting dynamic dimension size doesn't change the upper bound of the static
// shape.
let padded_v_five: f32[10] = set_dimension_size(v, five, /*dimension=*/0);
let padded_v_six: f32[10] = set_dimension_size(v, six, /*dimension=*/0);

// sum == 1 + 2 + 3 + 4 + 5
let sum:f32[] = reduce_sum(padded_v_five);
// product == 1 * 2 * 3 * 4 * 5
let product:f32[] = reduce_product(padded_v_five);

// Changing padding size will yield different result.
// sum == 1 + 2 + 3 + 4 + 5 + 6
let sum:f32[] = reduce_sum(padded_v_six);

GetTupleElement

See also XlaBuilder::GetTupleElement .

Indexes into a tuple with a compile-time-constant value.

The value must be a compile-time-constant so that shape inference can determine the type of the resulting value.

This is analogous to std::get<int N>(t) in C++. Conceptually:

let v: f32[10] = f32[10]{0, 1, 2, 3, 4, 5, 6, 7, 8, 9};
let s: s32 = 5;
let t: (f32[10], s32) = tuple(v, s);
let element_1: s32 = gettupleelement(t, 1);  // Inferred shape matches s32.

See also tf.tuple .

Infeed

See also XlaBuilder::Infeed .

Infeed(shape)

Argument Type Semantics
shape Shape Shape of the data read from the Infeed interface. The layout field of the shape must be set to match the layout of the data sent to the device; otherwise its behavior is undefined.

Reads a single data item from the implicit Infeed streaming interface of the device, interpreting the data as the given shape and its layout, and returns a XlaOp of the data. Multiple Infeed operations are allowed in a computation, but there must be a total order among the Infeed operations. For example, two Infeeds in the code below have a total order since there is a dependency between the while loops.

result1 = while (condition, init = init_value) {
  Infeed(shape)
}

result2 = while (condition, init = result1) {
  Infeed(shape)
}

Nested tuple shapes are not supported. For an empty tuple shape, the Infeed operation is effectively a no-op and proceeds without reading any data from the Infeed of the device.

Iota

See also XlaBuilder::Iota .

Iota(shape, iota_dimension)

Builds a constant literal on device rather than a potentially large host transfer. Creates an array that has specified shape and holds values starting at zero and incrementing by one along the specified dimension. For floating-point types, the produced array is equivalent to ConvertElementType(Iota(...)) where the Iota is of integral type and the conversion is to the floating-point type.

Arguments Type Semantics
shape Shape Shape of the array created by Iota()
iota_dimension int64 The dimension to increment along.

For example, Iota(s32[4, 8], 0) returns

  [[0, 0, 0, 0, 0, 0, 0, 0 ],
   [1, 1, 1, 1, 1, 1, 1, 1 ],
   [2, 2, 2, 2, 2, 2, 2, 2 ],
   [3, 3, 3, 3, 3, 3, 3, 3 ]]

Iota(s32[4, 8], 1) returns

  [[0, 1, 2, 3, 4, 5, 6, 7 ],
   [0, 1, 2, 3, 4, 5, 6, 7 ],
   [0, 1, 2, 3, 4, 5, 6, 7 ],
   [0, 1, 2, 3, 4, 5, 6, 7 ]]

Map

See also XlaBuilder::Map .

Map(operands..., computation)

Arguments Type Semantics
operands sequence of N XlaOp s N arrays of types T 0..T {N-1}
computation XlaComputation computation of type T_0, T_1, ..., T_{N + M -1} -> S with N parameters of type T and M of arbitrary type
dimensions int64 array array of map dimensions

Applies a scalar function over the given operands arrays, producing an array of the same dimensions where each element is the result of the mapped function applied to the corresponding elements in the input arrays.

The mapped function is an arbitrary computation with the restriction that it has N inputs of scalar type T and a single output with type S . The output has the same dimensions as the operands except that the element type T is replaced with S.

For example: Map(op1, op2, op3, computation, par1) maps elem_out <- computation(elem1, elem2, elem3, par1) at each (multi-dimensional) index in the input arrays to produce the output array.

OptimizationBarrier

Blocks any optimization pass from moving computations across the barrier.

Ensures that all inputs are evaluated before any operators that depend on the barrier's outputs.

Pad

See also XlaBuilder::Pad .

Pad(operand, padding_value, padding_config)

Arguments Type Semantics
operand XlaOp array of type T
padding_value XlaOp scalar of type T to fill in the added padding
padding_config PaddingConfig padding amount on both edges (low, high) and between the elements of each dimension

Expands the given operand array by padding around the array as well as between the elements of the array with the given padding_value . padding_config specifies the amount of edge padding and the interior padding for each dimension.

PaddingConfig is a repeated field of PaddingConfigDimension , which contains three fields for each dimension: edge_padding_low , edge_padding_high , and interior_padding .

edge_padding_low and edge_padding_high specify the amount of padding added at the low-end (next to index 0) and the high-end (next to the highest index) of each dimension respectively. The amount of edge padding can be negative -- the absolute value of negative padding indicates the number of elements to remove from the specified dimension.

interior_padding specifies the amount of padding added between any two elements in each dimension; it may not be negative. Interior padding occurs logically before edge padding, so in the case of negative edge padding, elements are removed from the interior-padded operand.

This operation is a no-op if the edge padding pairs are all (0, 0) and the interior padding values are all 0. The figure below shows examples of different edge_padding and interior_padding values for a two-dimensional array.

Recv

See also XlaBuilder::Recv .

Recv(shape, channel_handle)

Arguments Type Semantics
shape Shape shape of the data to receive
channel_handle ChannelHandle unique identifier for each send/recv pair

Receives data of the given shape from a Send instruction in another computation that shares the same channel handle. Returns a XlaOp for the received data.

The client API of Recv operation represents synchronous communication. However, the instruction is internally decomposed into 2 HLO instructions ( Recv and RecvDone ) to enable asynchronous data transfers. See also HloInstruction::CreateRecv and HloInstruction::CreateRecvDone .

Recv(const Shape& shape, int64 channel_id)

Allocates resources required to receive data from a Send instruction with the same channel_id. Returns a context for the allocated resources, which is used by a following RecvDone instruction to wait for the completion of the data transfer. The context is a tuple of {receive buffer (shape), request identifier (U32)} and it can only be used by a RecvDone instruction.

RecvDone(HloInstruction context)

Given a context created by a Recv instruction, waits for the data transfer to complete and returns the received data.

Reduce

See also XlaBuilder::Reduce .

Applies a reduction function to one or more arrays in parallel.

Reduce(operands..., init_values..., computation, dimensions)

Arguments Type Semantics
operands Sequence of N XlaOp N arrays of types T_0, ..., T_{N-1} .
init_values Sequence of N XlaOp N scalars of types T_0, ..., T_{N-1} .
computation XlaComputation computation of type T_0, ..., T_{N-1}, T_0, ..., T_{N-1} -> Collate(T_0, ..., T_{N-1}) .
dimensions int64 array unordered array of dimensions to reduce.

Where:

  • N is required to be greater or equal to 1.
  • The computation has to be "roughly" associative (see below).
  • All input arrays must have the same dimensions.
  • All initial values have to form an identity under computation .
  • If N = 1 , Collate(T) is T .
  • If N > 1 , Collate(T_0, ..., T_{N-1}) is a tuple of N elements of type T .

This operation reduces one or more dimensions of each input array into scalars. The rank of each returned array is rank(operand) - len(dimensions) . The output of the op is Collate(Q_0, ..., Q_N) where Q_i is an array of type T_i , the dimensions of which are described below.

Different backends are allowed to reassociate the reduction computation. This can lead to numerical differences, as some reduction functions like addition are not associative for floats. However, if the range of the data is limited, floating-point addition is close enough to being associative for most practical uses.

Examples

When reducing across one dimension in a single 1D array with values [10, 11, 12, 13] , with reduction function f (this is computation ) then that could be computed as

f(10, f(11, f(12, f(init_value, 13)))

but there are also many other possibilities, eg

f(init_value, f(f(10, f(init_value, 11)), f(f(init_value, 12), f(init_value, 13))))

The following is a rough pseudo-code example of how reduction could be implemented, using summation as the reduction computation with an initial value of 0.

result_shape <- remove all dims in dimensions from operand_shape

# Iterate over all elements in result_shape. The number of r's here is equal
# to the rank of the result
for r0 in range(result_shape[0]), r1 in range(result_shape[1]), ...:
  # Initialize this result element
  result[r0, r1...] <- 0

  # Iterate over all the reduction dimensions
  for d0 in range(dimensions[0]), d1 in range(dimensions[1]), ...:
    # Increment the result element with the value of the operand's element.
    # The index of the operand's element is constructed from all ri's and di's
    # in the right order (by construction ri's and di's together index over the
    # whole operand shape).
    result[r0, r1...] += operand[ri... di]

Here's an example of reducing a 2D array (matrix). The shape has rank 2, dimension 0 of size 2 and dimension 1 of size 3:

Results of reducing dimensions 0 or 1 with an "add" function:

Note that both reduction results are 1D arrays. The diagram shows one as column and another as row just for visual convenience.

For a more complex example, here is a 3D array. Its rank is 3, dimension 0 of size 4, dimension 1 of size 2 and dimension 2 of size 3. For simplicity, the values 1 to 6 are replicated across dimension 0.

Similarly to the 2D example, we can reduce just one dimension. If we reduce dimension 0, for example, we get a rank-2 array where all values across dimension 0 were folded into a scalar:

|  4   8  12 |
| 16  20  24 |

If we reduce dimension 2, we also get a rank-2 array where all values across dimension 2 were folded into a scalar:

| 6  15 |
| 6  15 |
| 6  15 |
| 6  15 |

Note that the relative order between the remaining dimensions in the input is preserved in the output, but some dimensions may get assigned new numbers (since the rank changes).

We can also reduce multiple dimensions. Add-reducing dimensions 0 and 1 produces the 1D array [20, 28, 36] .

Reducing the 3D array over all its dimensions produces the scalar 84 .

Variadic Reduce

When N > 1 , reduce function application is slightly more complex, as it is applied simultaneously to all inputs. The operands are supplied to the computation in the following order:

  • Running reduced value for the first operand
  • ...
  • Running reduced value for the N'th operand
  • Input value for the first operand
  • ...
  • Input value for the N'th operand

For example, consider the following reduction function, which can be used to compute the max and the argmax of a 1-D array in parallel:

f: (Float, Int, Float, Int) -> Float, Int
f(max, argmax, value, index):
  if value >= max:
    return (value, index)
  else:
    return (max, argmax)

For 1-D Input arrays V = Float[N], K = Int[N] , and init values I_V = Float, I_K = Int , the result f_(N-1) of reducing across the only input dimension is equivalent to the following recursive application:

f_0 = f(I_V, I_K, V_0, K_0)
f_1 = f(f_0.first, f_0.second, V_1, K_1)
...
f_(N-1) = f(f_(N-2).first, f_(N-2).second, V_(N-1), K_(N-1))

Applying this reduction to an array of values, and an array of sequential indices (ie iota), will co-iterate over the arrays, and return a tuple containing the maximal value and the matching index.

ReducePrecision

See also XlaBuilder::ReducePrecision .

Models the effect of converting floating-point values to a lower-precision format (such as IEEE-FP16) and back to the original format. The number of exponent and mantissa bits in the lower-precision format can be specified arbitrarily, although all bit sizes may not be supported on all hardware implementations.

ReducePrecision(operand, mantissa_bits, exponent_bits)

Arguments Type Semantics
operand XlaOp array of floating-point type T .
exponent_bits int32 number of exponent bits in lower-precision format
mantissa_bits int32 number of mantissa bits in lower-precision format

The result is an array of type T . The input values are rounded to the nearest value representable with the given number of mantissa bits (using "ties to even" semantics), and any values that exceed the range specified by the number of exponent bits are clamped to positive or negative infinity. NaN values are retained, although they may be converted to canonical NaN values.

The lower-precision format must have at least one exponent bit (in order to distinguish a zero value from an infinity, since both have a zero mantissa), and must have a non-negative number of mantissa bits. The number of exponent or mantissa bits may exceed the corresponding value for type T ; the corresponding portion of the conversion is then simply a no-op.

ReduceScatter

See also XlaBuilder::ReduceScatter .

ReduceScatter is a collective operation that effectively does an AllReduce and then scatters the result by splitting it into shard_count blocks along the scatter_dimension and replica i in the replica group receives the ith shard.

ReduceScatter(operand, computation, scatter_dim, shard_count, replica_group_ids, channel_id)

Arguments Type Semantics
operand XlaOp Array or a non-empty tuple of arrays to reduce across replicas.
computation XlaComputation Reduction computation
scatter_dimension int64 Dimension to scatter.
shard_count int64 Number of blocks to split scatter_dimension
replica_groups vector of vectors of int64 Groups between which the reductions are performed
channel_id optional int64 Optional channel ID for cross-module communication
  • When operand is a tuple of arrays, the reduce-scatter is performed on each element of the tuple.
  • replica_groups is a list of replica groups between which the reduction is performed (replica id for the current replica can be retrieved using ReplicaId ). The order of replicas in each group determines the order in which the all-reduce result will be scattered. replica_groups must either be empty (in which case all replicas belong to a single group), or contain the same number of elements as the number of replicas. When there are more than one replica groups, they all must be of the same size. For example, replica_groups = {0, 2}, {1, 3} performs reduction between the replicas 0 and 2 , and 1 and 3 and then scatters the result.
  • shard_count is the size of each replica group. We need this in cases where replica_groups are empty. If replica_groups is not empty, shard_count must be equal to the size of each replica group.
  • channel_id is used for cross-module communication: only reduce-scatter operations with the same channel_id can communicate with each other.

The output shape is the input shape with the scatter_dimension made shard_count times smaller. For example, if there are two replicas and the operand has the value [1.0, 2.25] and [3.0, 5.25] respectively on the two replicas, then the output value from this op where scatter_dim is 0 will be [4.0] for the first replica and [7.5] for the second replica.

ReduceWindow

See also XlaBuilder::ReduceWindow .

Applies a reduction function to all elements in each window of a sequence of N multi-dimensional arrays, producing a single or a tuple of N multi-dimensional arrays as output. Each output array has the same number of elements as the number of valid positions of the window. A pooling layer can be expressed as a ReduceWindow . Similar to Reduce , the applied computation is always passed the init_values on the left-hand side.

ReduceWindow(operands..., init_values..., computation, window_dimensions, window_strides, padding)

Arguments Type Semantics
operands N XlaOps A sequence of N multi-dimensional arrays of types T_0,..., T_{N-1} , each representing the base area on which the window is placed.
init_values N XlaOps The N starting values for the reduction, one for each of the N operands. See Reduce for details.
computation XlaComputation Reduction function of type T_0, ..., T_{N-1}, T_0, ..., T_{N-1} -> Collate(T_0, ..., T_{N-1}) , to apply to elements in each window of all the input operands.
window_dimensions ArraySlice<int64> array of integers for window dimension values
window_strides ArraySlice<int64> array of integers for window stride values
base_dilations ArraySlice<int64> array of integers for base dilation values
window_dilations ArraySlice<int64> array of integers for window dilation values
padding Padding padding type for window (Padding::kSame, which pads so as to have the same output shape as input if the stride is 1, or Padding::kValid, which uses no padding and "stops" the window once it no longer fits)

Where:

  • N is required to be greater or equal to 1.
  • All input arrays must have the same dimensions.
  • If N = 1 , Collate(T) is T .
  • If N > 1 , Collate(T_0, ..., T_{N-1}) is a tuple of N elements of type (T0,...T{N-1}) .

Below code and figure shows an example of using ReduceWindow . Input is a matrix of size [4x6] and both window_dimensions and window_stride_dimensions are [2x3].

// Create a computation for the reduction (maximum).
XlaComputation max;
{
  XlaBuilder builder(client_, "max");
  auto y = builder.Parameter(0, ShapeUtil::MakeShape(F32, {}), "y");
  auto x = builder.Parameter(1, ShapeUtil::MakeShape(F32, {}), "x");
  builder.Max(y, x);
  max = builder.Build().value();
}

// Create a ReduceWindow computation with the max reduction computation.
XlaBuilder builder(client_, "reduce_window_2x3");
auto shape = ShapeUtil::MakeShape(F32, {4, 6});
auto input = builder.Parameter(0, shape, "input");
builder.ReduceWindow(
    input,
    /*init_val=*/builder.ConstantLiteral(LiteralUtil::MinValue(F32)),
    *max,
    /*window_dimensions=*/{2, 3},
    /*window_stride_dimensions=*/{2, 3},
    Padding::kValid);

Stride of 1 in a dimension specifies that the position of a window in the dimension is 1 element away from its adjacent window. In order to specify that no windows overlap with each other, window_stride_dimensions should be equal to window_dimensions. The figure below illustrates the use of two different stride values. Padding is applied to each dimension of the input and the calculations are the same as though the input came in with the dimensions it has after padding.

For a non-trivial padding example, consider computing reduce-window minimum (initial value is MAX_FLOAT ) with dimension 3 and stride 2 over the input array [10000, 1000, 100, 10, 1] . Padding kValid computes minimums over two valid windows: [10000, 1000, 100] and [100, 10, 1] , resulting in the output [100, 1] . Padding kSame first pads the array so that the shape after the reduce-window would be the same as input for stride one by adding initial elements on both sides, getting [MAX_VALUE, 10000, 1000, 100, 10, 1, MAX_VALUE] . Running reduce-window over the padded array operates on three windows [MAX_VALUE, 10000, 1000] , [1000, 100, 10] , [10, 1, MAX_VALUE] , and yields [1000, 10, 1] .

The evaluation order of the reduction function is arbitrary and may be non-deterministic. Therefore, the reduction function should not be overly sensitive to reassociation. See the discussion about associativity in the context of Reduce for more details.

ReplicaId

See also XlaBuilder::ReplicaId .

Returns the unique ID (U32 scalar) of the replica.

ReplicaId()

The unique ID of each replica is an unsigned integer in the interval [0, N) , where N is the number of replicas. Since all the replicas are running the same program, a ReplicaId() call in the program will return a different value on each replica.

Reshape

See also XlaBuilder::Reshape and the Collapse operation.

Reshapes the dimensions of an array into a new configuration.

Reshape(operand, new_sizes) Reshape(operand, dimensions, new_sizes)

Arguments Type Semantics
operand XlaOp array of type T
dimensions int64 vector order in which dimensions are collapsed
new_sizes int64 vector vector of sizes of new dimensions

Conceptually, reshape first flattens an array into a one-dimensional vector of data values, and then refines this vector into a new shape. The input arguments are an arbitrary array of type T, a compile-time-constant vector of dimension indices, and a compile-time-constant vector of dimension sizes for the result. The values in the dimension vector, if given, must be a permutation of all of T's dimensions; the default if not given is {0, ..., rank - 1} . The order of the dimensions in dimensions is from slowest-varying dimension (most major) to fastest-varying dimension (most minor) in the loop nest which collapses the input array into a single dimension. The new_sizes vector determines the size of the output array. The value at index 0 in new_sizes is the size of dimension 0, the value at index 1 is the size of dimension 1, and so on. The product of the new_size dimensions must equal the product of the operand's dimension sizes. When refining the collapsed array into the multidimensional array defined by new_sizes , the dimensions in new_sizes are ordered from slowest varying (most major) and to fastest varying (most minor).

For example, let v be an array of 24 elements:

let v = f32[4x2x3] { { {10, 11, 12}, {15, 16, 17} },
                    { {20, 21, 22}, {25, 26, 27} },
                    { {30, 31, 32}, {35, 36, 37} },
                    { {40, 41, 42}, {45, 46, 47} } };

In-order collapse:
let v012_24 = Reshape(v, {0,1,2}, {24});
then v012_24 == f32[24] {10, 11, 12, 15, 16, 17, 20, 21, 22, 25, 26, 27,
                         30, 31, 32, 35, 36, 37, 40, 41, 42, 45, 46, 47};

let v012_83 = Reshape(v, {0,1,2}, {8,3});
then v012_83 == f32[8x3] { {10, 11, 12}, {15, 16, 17},
                          {20, 21, 22}, {25, 26, 27},
                          {30, 31, 32}, {35, 36, 37},
                          {40, 41, 42}, {45, 46, 47} };

Out-of-order collapse:
let v021_24 = Reshape(v, {1,2,0}, {24});
then v012_24 == f32[24]  {10, 20, 30, 40, 11, 21, 31, 41, 12, 22, 32, 42,
                          15, 25, 35, 45, 16, 26, 36, 46, 17, 27, 37, 47};

let v021_83 = Reshape(v, {1,2,0}, {8,3});
then v021_83 == f32[8x3] { {10, 20, 30}, {40, 11, 21},
                          {31, 41, 12}, {22, 32, 42},
                          {15, 25, 35}, {45, 16, 26},
                          {36, 46, 17}, {27, 37, 47} };


let v021_262 = Reshape(v, {1,2,0}, {2,6,2});
then v021_262 == f32[2x6x2] { { {10, 20}, {30, 40},
                              {11, 21}, {31, 41},
                              {12, 22}, {32, 42} },
                             { {15, 25}, {35, 45},
                              {16, 26}, {36, 46},
                              {17, 27}, {37, 47} } };

As a special case, reshape can transform a single-element array to a scalar and vice versa. For example,

Reshape(f32[1x1] { {5} }, {0,1}, {}) == 5;
Reshape(5, {}, {1,1}) == f32[1x1] { {5} };

Rev (reverse)

See also XlaBuilder::Rev .

Rev(operand, dimensions)

Arguments Type Semantics
operand XlaOp array of type T
dimensions ArraySlice<int64> dimensions to reverse

Reverses the order of elements in the operand array along the specified dimensions , generating an output array of the same shape. Each element of the operand array at a multidimensional index is stored into the output array at a transformed index. The multidimensional index is transformed by reversing the index in each dimension to be reversed (ie, if a dimension of size N is one of the reversing dimensions, its index i is transformed into N - 1 - i).

One use for the Rev operation is to reverse the convolution weight array along the two window dimensions during the gradient computation in neural networks.

RngNormal

See also XlaBuilder::RngNormal .

Constructs an output of a given shape with random numbers generated following the \(N(\mu, \sigma)\) normal distribution. The parameters \(\mu\) and\(\sigma\), and output shape have to have a floating point elemental type. The parameters furthermore have to be scalar valued.

RngNormal(mu, sigma, shape)

Arguments Type Semantics
mu XlaOp Scalar of type T specifying mean of generated numbers
sigma XlaOp Scalar of type T specifying standard deviation of generated numbers
shape Shape Output shape of type T

RngUniform

See also XlaBuilder::RngUniform .

Constructs an output of a given shape with random numbers generated following the uniform distribution over the interval \([a,b)\). The parameters and output element type have to be a boolean type, an integral type or a floating point types, and the types have to be consistent. The CPU and GPU backends currently only support F64, F32, F16, BF16, S64, U64, S32 and U32. Furthermore, the parameters need to be scalar valued. If \(b <= a\) the result is implementation-defined.

RngUniform(a, b, shape)

Arguments Type Semantics
a XlaOp Scalar of type T specifying lower limit of interval
b XlaOp Scalar of type T specifying upper limit of interval
shape Shape Output shape of type T

RngBitGenerator

Generates an output with a given shape filled with uniform random bits using the specified algorithm (or backend default) and returns an updated state (with the same shape as initial state) and the generated random data.

Initial state is the initial state of the current random number generation. It and the required shape and valid values are dependent on the algorithm used.

The output is guaranteed to be a deterministic function of the initial state but it is not guaranteed to be deterministic between backends and different compiler versions.

RngBitGenerator(algorithm, key, shape)

Arguments Type Semantics
algorithm RandomAlgorithm PRNG algorithm to be used.
initial_state XlaOp Initial state for the PRNG algorithm.
shape Shape Output shape for generated data.

Available values for algorithm :

Scatter

The XLA scatter operation generates a sequence of results which are the values of the input array operands , with several slices (at indices specified by scatter_indices ) updated with the sequence of values in updates using update_computation .

See also XlaBuilder::Scatter .

scatter(operands..., scatter_indices, updates..., update_computation, index_vector_dim, update_window_dims, inserted_window_dims, scatter_dims_to_operand_dims)

Arguments Type Semantics
operands Sequence of N XlaOp N arrays of types T_0, ..., T_N to be scattered into.
scatter_indices XlaOp Array containing the starting indices of the slices that must be scattered to.
updates Sequence of N XlaOp N arrays of types T_0, ..., T_N . updates[i] contains the values that must be used for scattering operands[i] .
update_computation XlaComputation Computation to be used for combining the existing values in the input array and the updates during scatter. This computation should be of type T_0, ..., T_N, T_0, ..., T_N -> Collate(T_0, ..., T_N) .
index_vector_dim int64 The dimension in scatter_indices that contains the starting indices.
update_window_dims ArraySlice<int64> The set of dimensions in updates shape that are window dimensions .
inserted_window_dims ArraySlice<int64> The set of window dimensions that must be inserted into updates shape.
scatter_dims_to_operand_dims ArraySlice<int64> A dimensions map from the scatter indices to the operand index space. This array is interpreted as mapping i to scatter_dims_to_operand_dims[i] . It has to be one-to-one and total.
indices_are_sorted bool Whether the indices are guaranteed to be sorted by the caller.

Where:

  • N is required to be greater or equal to 1.
  • operands [ 0 ], ..., operands [ N-1 ] must all have the same dimensions.
  • updates [ 0 ], ..., updates [ N-1 ] must all have the same dimensions.
  • If N = 1 , Collate(T) is T .
  • If N > 1 , Collate(T_0, ..., T_N) is a tuple of N elements of type T .

If index_vector_dim is equal to scatter_indices.rank we implicitly consider scatter_indices to have a trailing 1 dimension.

We define update_scatter_dims of type ArraySlice<int64> as the set of dimensions in updates shape that are not in update_window_dims , in ascending order.

The arguments of scatter should follow these constraints:

  • Each updates array must be of rank update_window_dims.size + scatter_indices.rank - 1 .

  • Bounds of dimension i in each updates array must conform to the following:

    • If i is present in update_window_dims (ie equal to update_window_dims [ k ] for some k ), then the bound of dimension i in updates must not exceed the corresponding bound of operand after accounting for the inserted_window_dims (ie adjusted_window_bounds [ k ], where adjusted_window_bounds contains the bounds of operand with the bounds at indices inserted_window_dims removed).
    • If i is present in update_scatter_dims (ie equal to update_scatter_dims [ k ] for some k ), then the bound of dimension i in updates must be equal to the corresponding bound of scatter_indices , skipping index_vector_dim (ie scatter_indices.shape.dims [ k ], if k < index_vector_dim and scatter_indices.shape.dims [ k+1 ] otherwise).
  • update_window_dims must be in ascending order, not have any repeating dimension numbers, and be in the range [0, updates.rank) .

  • inserted_window_dims must be in ascending order, not have any repeating dimension numbers, and be in the range [0, operand.rank) .

  • operand.rank must equal the sum of update_window_dims.size and inserted_window_dims.size .

  • scatter_dims_to_operand_dims.size must be equal to scatter_indices.shape.dims [ index_vector_dim ], and its values must be in the range [0, operand.rank) .

For a given index U in each updates array, the corresponding index I in the corresponding operands array into which this update has to be applied is computed as follows:

  1. Let G = { U [ k ] for k in update_scatter_dims }. Use G to look up an index vector S in the scatter_indices array such that S [ i ] = scatter_indices [Combine( G , i )] where Combine(A, b) inserts b at positions index_vector_dim into A.
  2. Create an index S in into operand using S by scattering S using the scatter_dims_to_operand_dims map. More formally:
    1. S in [ scatter_dims_to_operand_dims [ k ]] = S [ k ] if k < scatter_dims_to_operand_dims.size .
    2. S in [ _ ] = 0 otherwise.
  3. Create an index W in into each operands array by scattering the indices at update_window_dims in U according to inserted_window_dims . More formally:
    1. W in [ window_dims_to_operand_dims ( k )] = U [ k ] if k is in update_window_dims , where window_dims_to_operand_dims is the monotonic function with domain [ 0 , update_window_dims.size ) and range [ 0 , operand.rank ) \ inserted_window_dims . (For example, if update_window_dims.size is 4 , operand.rank is 6 , and inserted_window_dims is { 0 , 2 } then window_dims_to_operand_dims is { 01 , 13 , 24 , 35 }).
    2. W in [ _ ] = 0 otherwise.
  4. I is W in + S in where + is element-wise addition.

In summary, the scatter operation can be defined as follows.

  • Initialize output with operands , ie for all indices J , for all indices O in the operands [ J ] array:
    output [ J ][ O ] = operands [ J ][ O ]
  • For every index U in the updates [ J ] array and the corresponding index O in the operand [ J ] array, if O is a valid index for output :
    (output [ 0 ][ O ], ..., output [ N-1 ][ O ]) = update_computation ( output [ 0 ][ O ], ..., , output [ N-1 ][ O ], updates [ 0 ][ U ], ..., updates [ N-1 ][ U ])

The order in which updates are applied is non-deterministic. So, when multiple indices in updates refer to the same index in operands , the corresponding value in output will be non-deterministic.

Note that the first parameter that is passed into the update_computation will always be the current value from the output array and the second parameter will always be the value from the updates array. This is important specifically for cases when the update_computation is not commutative .

If indices_are_sorted is set to true then XLA can assume that start_indices are sorted (in ascending start_index_map order) by the user. If they are not then the semantics is implementation defined.

Informally, the scatter op can be viewed as an inverse of the gather op, ie the scatter op updates the elements in the input that are extracted by the corresponding gather op.

For a detailed informal description and examples, refer to the "Informal Description" section under Gather .

Select

See also XlaBuilder::Select .

Constructs an output array from elements of two input arrays, based on the values of a predicate array.

Select(pred, on_true, on_false)

Arguments Type Semantics
pred XlaOp array of type PRED
on_true XlaOp array of type T
on_false XlaOp array of type T

The arrays on_true and on_false must have the same shape. This is also the shape of the output array. The array pred must have the same dimensionality as on_true and on_false , with the PRED element type.

For each element P of pred , the corresponding element of the output array is taken from on_true if the value of P is true , and from on_false if the value of P is false . As a restricted form of broadcasting , pred can be a scalar of type PRED . In this case, the output array is taken wholly from on_true if pred is true , and from on_false if pred is false .

Example with non-scalar pred :

let pred: PRED[4] = {true, false, false, true};
let v1: s32[4] = {1, 2, 3, 4};
let v2: s32[4] = {100, 200, 300, 400};
==>
Select(pred, v1, v2) = s32[4]{1, 200, 300, 4};

Example with scalar pred :

let pred: PRED = true;
let v1: s32[4] = {1, 2, 3, 4};
let v2: s32[4] = {100, 200, 300, 400};
==>
Select(pred, v1, v2) = s32[4]{1, 2, 3, 4};

Selections between tuples are supported. Tuples are considered to be scalar types for this purpose. If on_true and on_false are tuples (which must have the same shape!) then pred has to be a scalar of type PRED .

SelectAndScatter

See also XlaBuilder::SelectAndScatter .

This operation can be considered as a composite operation that first computes ReduceWindow on the operand array to select an element from each window, and then scatters the source array to the indices of the selected elements to construct an output array with the same shape as the operand array. The binary select function is used to select an element from each window by applying it across each window, and it is called with the property that the first parameter's index vector is lexicographically less than the second parameter's index vector. The select function returns true if the first parameter is selected and returns false if the second parameter is selected, and the function must hold transitivity (ie, if select(a, b) and select(b, c) are true , then select(a, c) is also true ) so that the selected element does not depend on the order of the elements traversed for a given window.

The function scatter is applied at each selected index in the output array. It takes two scalar parameters:

  1. Current value at the selected index in the output array
  2. The scatter value from source that applies to the selected index

It combines the two parameters and returns a scalar value that's used to update the value at the selected index in the output array. Initially, all indices of the output array are set to init_value .

The output array has the same shape as the operand array and the source array must have the same shape as the result of applying a ReduceWindow operation on the operand array. SelectAndScatter can be used to backpropagate the gradient values for a pooling layer in a neural network.

SelectAndScatter(operand, select, window_dimensions, window_strides, padding, source, init_value, scatter)

Arguments Type Semantics
operand XlaOp array of type T over which the windows slide
select XlaComputation binary computation of type T, T -> PRED , to apply to all elements in each window; returns true if the first parameter is selected and returns false if the second parameter is selected
window_dimensions ArraySlice<int64> array of integers for window dimension values
window_strides ArraySlice<int64> array of integers for window stride values
padding Padding padding type for window (Padding::kSame or Padding::kValid)
source XlaOp array of type T with the values to scatter
init_value XlaOp scalar value of type T for the initial value of the output array
scatter XlaComputation binary computation of type T, T -> T , to apply each scatter source element with its destination element

The figure below shows examples of using SelectAndScatter , with the select function computing the maximal value among its parameters. Note that when the windows overlap, as in the figure (2) below, an index of the operand array may be selected multiple times by different windows. In the figure, the element of value 9 is selected by both of the top windows (blue and red) and the binary addition scatter function produces the output element of value 8 (2 + 6).

The evaluation order of the scatter function is arbitrary and may be non-deterministic. Therefore, the scatter function should not be overly sensitive to reassociation. See the discussion about associativity in the context of Reduce for more details.

Send

See also XlaBuilder::Send .

Send(operand, channel_handle)

Arguments Type Semantics
operand XlaOp data to send (array of type T)
channel_handle ChannelHandle unique identifier for each send/recv pair

Sends the given operand data to a Recv instruction in another computation that shares the same channel handle. Does not return any data.

Similar to the Recv operation, the client API of Send operation represents synchronous communication, and is internally decomposed into 2 HLO instructions ( Send and SendDone ) to enable asynchronous data transfers. See also HloInstruction::CreateSend and HloInstruction::CreateSendDone .

Send(HloInstruction operand, int64 channel_id)

Initiates an asynchronous transfer of the operand to the resources allocated by the Recv instruction with the same channel id. Returns a context, which is used by a following SendDone instruction to wait for the completion of the data transfer. The context is a tuple of {operand (shape), request identifier (U32)} and it can only be used by a SendDone instruction.

SendDone(HloInstruction context)

Given a context created by a Send instruction, waits for the data transfer to complete. The instruction does not return any data.

Scheduling of channel instructions

The execution order of the 4 instructions for each channel ( Recv , RecvDone , Send , SendDone ) is as below.

  • Recv happens before Send
  • Send happens before RecvDone
  • Recv happens before RecvDone
  • Send happens before SendDone

When the backend compilers generate a linear schedule for each computation that communicates via channel instructions, there must not be cycles across the computations. For example, below schedules lead to deadlocks.

Slice

See also XlaBuilder::Slice .

Slicing extracts a sub-array from the input array. The sub-array is of the same rank as the input and contains the values inside a bounding box within the input array where the dimensions and indices of the bounding box are given as arguments to the slice operation.

Slice(operand, start_indices, limit_indices, strides)

Arguments Type Semantics
operand XlaOp N dimensional array of type T
start_indices ArraySlice<int64> List of N integers containing the starting indices of the slice for each dimension. Values must be greater than or equal to zero.
limit_indices ArraySlice<int64> List of N integers containing the ending indices (exclusive) for the slice for each dimension. Each value must be greater than or equal to the respective start_indices value for the dimension and less than or equal to the size of the dimension.
strides ArraySlice<int64> List of N integers that decides the input stride of the slice. The slice picks every strides[d] element in dimension d .

1-dimensional example:

let a = {0.0, 1.0, 2.0, 3.0, 4.0}
Slice(a, {2}, {4}) produces:
  {2.0, 3.0}

2-dimensional example:

let b =
 { {0.0,  1.0,  2.0},
   {3.0,  4.0,  5.0},
   {6.0,  7.0,  8.0},
   {9.0, 10.0, 11.0} }

Slice(b, {2, 1}, {4, 3}) produces:
  { { 7.0,  8.0},
    {10.0, 11.0} }

Sort

See also XlaBuilder::Sort .

Sort(operands, comparator, dimension, is_stable)

Arguments Type Semantics
operands ArraySlice<XlaOp> The operands to sort.
comparator XlaComputation The comparator computation to use.
dimension int64 The dimension along which to sort.
is_stable bool Whether stable sorting should be used.

If only one operand is provided:

  • If the operand is a rank-1 tensor (an array), the result is a sorted array. If you want to sort the array into ascending order, the comparator should perform a less-than comparison. Formally, after the array is sorted, it holds for all index positions i, j with i < j that either comparator(value[i], value[j]) = comparator(value[j], value[i]) = false or comparator(value[i], value[j]) = true .

  • If the operand has higher rank, the operand is sorted along the provided dimension. For example, for a rank-2 tensor (a matrix), a dimension value of 0 will independently sort every column, and a dimension value of 1 will independently sort each row. If no dimension number is provided, then the last dimension is chosen by default. For the dimension which is sorted, the same sorting order applies as in the rank-1 case.

If n > 1 operands are provided:

  • All n operands must be tensors with the same dimensions. The element types of the tensors may be different.

  • All operands are sorted together, not individually. Conceptually the operands are treated as a tuple. When checking whether the elements of each operand at index positions i and j need to be swapped, the comparator is called with 2 * n scalar parameters, where parameter 2 * k corresponds to the value at position i from the k-th operand, and parameter 2 * k + 1 corresponds to the value at position j from the k-th operand. Usually, the comparator would thus compare parameters 2 * k and 2 * k + 1 with each other and possibly use other parameter pairs as tie breakers.

  • The result is a tuple that consists of the operands in sorted order (along the provided dimension, as above). The i-th operand of the tuple corresponds to the i-th operand of Sort.

For example, if there are three operands operand0 = [3, 1] , operand1 = [42, 50] , operand2 = [-3.0, 1.1] , and the comparator compares only the values of operand0 with less-than, then the output of the sort is the tuple ([1, 3], [50, 42], [1.1, -3.0]) .

If is_stable is set to true, the sort is guaranteed to be stable, that is, if there are elements which are considered to be equal by the comparator, the relative order of the equal values is preserved. Two elements e1 and e2 are equal if and only if comparator(e1, e2) = comparator(e2, e1) = false . By default, is_stable is set to false.

Top-K

See also the jax.lax.top_k operation.

TopK(operand)

Arguments Type Semantics
operand XlaOp N-dimensional array
k int64 Integer specifying the number of top entries.
comparator XlaComputation The comparator computation to use.

Returns top k values and their indices as a tuple, along the last dimension of the operand using the given comparator (for usual topk behavior, it should be strict-greater-than operation).

For example, given strict > operator, k=1 and the following operand of shape f32[2,3] :

[[0.1, 0.3, 0.1], [0.7, 0.2, -0.1]]

The TopK application returns the following tuple of shape (f32[2,1], s32[2,1]) :

([[0.3], [0.7]], [[1], [0]])

Transpose

See also the tf.reshape operation.

Transpose(operand)

Arguments Type Semantics
operand XlaOp The operand to transpose.
permutation ArraySlice<int64> How to permute the dimensions.

Permutes the operand dimensions with the given permutation, so ∀ i . 0 ≤ i < rank ⇒ input_dimensions[permutation[i]] = output_dimensions[i] .

This is the same as Reshape(operand, permutation, Permute(permutation, operand.shape.dimensions)).

TriangularSolve

See also XlaBuilder::TriangularSolve .

Solves systems of linear equations with lower or upper triangular coefficient matrices by forward- or back-substitution. Broadcasting along leading dimensions, this routine solves one of the matrix systems op(a) * x = b , or x * op(a) = b , for the variable x , given a and b , where op(a) is either op(a) = a , or op(a) = Transpose(a) , or op(a) = Conj(Transpose(a)) .

TriangularSolve(a, b, left_side, lower, unit_diagonal, transpose_a)

Arguments Type Semantics
a XlaOp a rank > 2 array of a complex or floating-point type with shape [..., M, M] .
b XlaOp a rank > 2 array of the same type with shape [..., M, K] if left_side is true, [..., K, M] otherwise.
left_side bool indicates whether to solve a system of the form op(a) * x = b ( true ) or x * op(a) = b ( false ).
lower bool whether to use the upper or lower triangle of a .
unit_diagonal bool if true , the diagonal elements of a are assumed to be 1 and not accessed.
transpose_a Transpose whether to use a as is, transpose it or take its conjugate transpose.

Input data is read only from the lower/upper triangle of a , depending on the value of lower . Values from the other triangle are ignored. Output data is returned in the same triangle; the values in the other triangle are implementation-defined and may be anything.

If the rank of a and b are greater than 2, they are treated as batches of matrices, where all except the minor 2 dimensions are batch dimensions. a and b must have equal batch dimensions.

Tuple

See also XlaBuilder::Tuple .

A tuple containing a variable number of data handles, each of which has its own shape.

This is analogous to std::tuple in C++. Conceptually:

let v: f32[10] = f32[10]{0, 1, 2, 3, 4, 5, 6, 7, 8, 9};
let s: s32 = 5;
let t: (f32[10], s32) = tuple(v, s);

Tuples can be deconstructed (accessed) via the GetTupleElement operation.

While

See also XlaBuilder::While .

While(condition, body, init)

Arguments Type Semantics
condition XlaComputation XlaComputation of type T -> PRED which defines the termination condition of the loop.
body XlaComputation XlaComputation of type T -> T which defines the body of the loop.
init T Initial value for the parameter of condition and body .

Sequentially executes the body until the condition fails. This is similar to a typical while loop in many other languages except for the differences and restrictions listed below.

  • A While node returns a value of type T , which is the result from the last execution of the body .
  • The shape of the type T is statically determined and must be the same across all iterations.

The T parameters of the computations are initialized with the init value in the first iteration and are automatically updated to the new result from body in each subsequent iteration.

One main use case of the While node is to implement the repeated execution of training in neural networks. Simplified pseudocode is shown below with a graph that represents the computation. The code can be found in while_test.cc . The type T in this example is a Tuple consisting of an int32 for the iteration count and a vector[10] for the accumulator. For 1000 iterations, the loop keeps adding a constant vector to the accumulator.

// Pseudocode for the computation.
init = {0, zero_vector[10]} // Tuple of int32 and float[10].
result = init;
while (result(0) < 1000) {
  iteration = result(0) + 1;
  new_vector = result(1) + constant_vector[10];
  result = {iteration, new_vector};
}