در زیر معنای عملیات تعریف شده در رابط XlaBuilder
را شرح می دهد. به طور معمول، این عملیات یک به یک به عملیات تعریف شده در رابط RPC در xla_data.proto
نگاشت.
نکته ای در مورد نامگذاری: نوع داده تعمیم یافته ای که XLA با آن سروکار دارد، یک آرایه N بعدی است که عناصری از نوع یکنواخت (مانند شناور ۳۲ بیتی) را در خود نگه می دارد. در سراسر مستندات، آرایه برای نشان دادن یک آرایه با ابعاد دلخواه استفاده می شود. برای راحتی، موارد خاص نام های خاص و آشناتری دارند. به عنوان مثال یک بردار یک آرایه 1 بعدی و یک ماتریس یک آرایه دو بعدی است.
گذشته از همه اینها
XlaBuilder::AfterAll
نیز ببینید.
AfterAll تعداد متغیری توکن را می گیرد و یک توکن تولید می کند. توکنها انواع اولیهای هستند که میتوان آنها را بین عملیاتهای جانبی قرار داد تا سفارش را اجرا کنند. AfterAll
می توان به عنوان پیوندی از نشانه ها برای سفارش عملیات پس از یک مجموعه عملیات استفاده کرد.
AfterAll(operands)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operands | XlaOp | تعداد متغیر توکن ها |
همه جمع شوند
XlaBuilder::AllGather
نیز ببینید.
الحاق بین کپی ها را انجام می دهد.
AllGather(operand, all_gather_dim, shard_count, replica_group_ids, channel_id)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه ای برای الحاق بین کپی ها. |
all_gather_dim | int64 | بعد الحاق. |
replica_groups | بردار بردارهای int64 | گروه هایی که الحاق بین آنها انجام می شود. |
channel_id | اختیاری int64 | شناسه کانال اختیاری برای ارتباط متقابل ماژول. |
-
replica_groups
لیستی از گروههای replica است که الحاق بین آنها انجام میشود (شناسه replica برای ماکت فعلی را میتوان با استفاده ازReplicaId
بازیابی کرد). ترتیب تکرارها در هر گروه تعیین کننده ترتیب قرار گرفتن ورودی های آنها در نتیجه است.replica_groups
یا باید خالی باشند (در این صورت همه کپیها متعلق به یک گروه واحد هستند که از0
تاN - 1
مرتب شدهاند)، یا دارای همان تعداد عناصر با تعداد کپیها باشند. برای مثال،replica_groups = {0, 2}, {1, 3}
الحاق بین replica های0
و2
و1
و3
را انجام می دهد. -
shard_count
اندازه هر گروه ماکت است. در مواردی کهreplica_groups
خالی هستند به این نیاز داریم. -
channel_id
برای ارتباط متقابل ماژول استفاده می شود: فقط عملیاتall-gather
با همانchannel_id
می توانند با یکدیگر ارتباط برقرار کنند.
شکل خروجی شکل ورودی با all_gather_dim
است که shard_count
بار بزرگتر شده است. به عنوان مثال، اگر دو replica وجود داشته باشد و عملوند دارای مقدار [1.0, 2.5]
و [3.0, 5.25]
به ترتیب روی دو replica باشد، آنگاه مقدار خروجی از این عملیات که در آن all_gather_dim
0
است [1.0, 2.5, 3.0, 5.25]
خواهد بود. [1.0, 2.5, 3.0, 5.25]
در هر دو ماکت.
همه کاهش
XlaBuilder::AllReduce
نیز ببینید.
یک محاسبات سفارشی در بین کپی ها انجام می دهد.
AllReduce(operand, computation, replica_group_ids, channel_id)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه یا چند تایی از آرایههای خالی برای کاهش بین ماکتها. |
computation | XlaComputation | محاسبه کاهش |
replica_groups | بردار بردارهای int64 | گروه هایی که بین آنها کاهش ها انجام می شود |
channel_id | اختیاری int64 | شناسه کانال اختیاری برای ارتباط متقابل ماژول |
- هنگامی که
operand
چند آرایه است، کاهش همه روی هر عنصر تاپل انجام می شود. -
replica_groups
فهرستی از گروههای replica است که بین آنها کاهش انجام میشود (شناسه replica برای replica فعلی را میتوان با استفاده ازReplicaId
بازیابی کرد).replica_groups
یا باید خالی باشد (در این صورت همه replica ها متعلق به یک گروه واحد هستند)، یا دارای همان تعداد عناصر با تعداد Replica ها باشند. برای مثال،replica_groups = {0, 2}, {1, 3}
بین کپیهای0
و2
و1
و3
کاهش میدهد. -
channel_id
برای ارتباطات متقابل ماژول استفاده میشود: فقط عملیاتهایall-reduce
با همانchannel_id
میتوانند با یکدیگر ارتباط برقرار کنند.
شکل خروجی همان شکل ورودی است. به عنوان مثال، اگر دو replica وجود داشته باشد و عملوند دارای مقدار [1.0, 2.5]
و [3.0, 5.25]
به ترتیب روی دو replica باشد، آنگاه مقدار خروجی از این عملیات و محاسبه جمع در هر دو [4.0, 7.75]
خواهد بود. ماکت ها اگر ورودی یک تاپلی باشد، خروجی نیز یک تاپل است.
محاسبه نتیجه AllReduce
مستلزم داشتن یک ورودی از هر ماکت است، بنابراین اگر یک ماکت یک گره AllReduce
را بیشتر از دیگری اجرا کند، آنگاه ماکت قبلی برای همیشه منتظر خواهد ماند. از آنجایی که کپیها همه یک برنامه را اجرا میکنند، راههای زیادی برای این کار وجود ندارد، اما زمانی ممکن است که شرایط حلقه while به دادههای ورودی بستگی داشته باشد و دادههایی که وارد میشوند باعث شوند حلقه while بارها تکرار شود. روی یک ماکت نسبت به دیگری
AllToAll
XlaBuilder::AllToAll
نیز ببینید.
AllToAll یک عملیات جمعی است که داده ها را از همه هسته ها به همه هسته ها ارسال می کند. دو فاز دارد:
- فاز پراکندگی. در هر هسته، عملوند به تعداد بلوکهای
split_count
در امتدادsplit_dimensions
تقسیم میشود، و بلوکها به همه هستهها پراکنده میشوند، به عنوان مثال، بلوک ith به هسته ith ارسال میشود. - مرحله جمع آوری هر هسته بلوک های دریافتی را در امتداد
concat_dimension
به هم متصل می کند.
هسته های شرکت کننده را می توان توسط:
-
replica_groups
: هر ReplicaGroup حاوی لیستی از Replica id شرکت کننده در محاسبات است (شناسه replica برای replica فعلی را می توان با استفاده ازReplicaId
بازیابی کرد). AllToAll در زیر گروه ها به ترتیب مشخص شده اعمال خواهد شد. برای مثال،replica_groups = { {1,2,3}, {4,5,0} }
به این معنی است که یک AllToAll در کپیهای{1, 2, 3}
و در مرحله جمعآوری اعمال میشود و بلوکهای دریافتی به همان ترتیب 1، 2، 3 الحاق شود. سپس، AllToAll دیگری در کپی های 4، 5، 0 اعمال می شود، و ترتیب الحاق نیز 4، 5، 0 است. اگرreplica_groups
خالی باشد، همه کپی ها متعلق به یک هستند. گروه، به ترتیب الحاق ظاهرشان.
پیش نیازها:
- اندازه بعد عملوند در
split_dimension
برsplit_count
قابل تقسیم است. - شکل عملوند تاپلی نیست.
AllToAll(operand, split_dimension, concat_dimension, split_count, replica_groups)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | n آرایه ورودی بعدی |
split_dimension | int64 | مقداری در بازه [0, n) که ابعادی را که عملوند در امتداد آن تقسیم میشود نامگذاری میکند |
concat_dimension | int64 | مقداری در بازه [0, n) که ابعادی را که بلوک های تقسیم شده در امتداد آن به هم متصل می شوند نام می برد. |
split_count | int64 | تعداد هسته هایی که در این عملیات شرکت می کنند. اگر replica_groups خالی است، این باید تعداد replica ها باشد. در غیر این صورت، این باید برابر با تعداد کپی در هر گروه باشد. |
replica_groups | وکتور ReplicaGroup | هر گروه شامل لیستی از شناسه ماکت است. |
در زیر نمونه ای از Alltoall را نشان می دهد.
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);

در این مثال، 4 هسته در Alltoall شرکت می کنند. در هر هسته، عملوند در امتداد بعد 1 به 4 قسمت تقسیم می شود، بنابراین هر قسمت دارای شکل f32 [4،4] است. 4 قسمت در تمام هسته ها پراکنده شده اند. سپس هر هسته قطعات دریافتی را در امتداد ابعاد 0 به ترتیب یا هسته 0-4 به هم متصل می کند. بنابراین خروجی هر هسته دارای شکل f32 [16,4] است.
BatchNormGrad
همچنین XlaBuilder::BatchNormGrad
و مقاله نرمال سازی دسته اصلی را برای توضیح دقیق الگوریتم ببینید.
گرادیان های هنجار دسته ای را محاسبه می کند.
BatchNormGrad(operand, scale, mean, variance, grad_output, epsilon, feature_index)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | n آرایه بعدی باید نرمال شود (x) |
scale | XlaOp | آرایه 1 بعدی (\(\gamma\)) |
mean | XlaOp | آرایه 1 بعدی (\(\mu\)) |
variance | XlaOp | آرایه 1 بعدی (\(\sigma^2\)) |
grad_output | XlaOp | گرادیان ها به BatchNormTraining (\( \nabla y\)) منتقل شد |
epsilon | float | مقدار اپسیلون (\(\epsilon\)) |
feature_index | int64 | شاخص به بعد ویژگی در operand |
برای هر ویژگی در بعد ویژگی ( feature_index
شاخص بعد ویژگی در operand
است)، این عملیات گرادیان ها را با توجه به operand
، offset
و scale
در تمام ابعاد دیگر محاسبه می کند. feature_index
باید یک شاخص معتبر برای بعد ویژگی در operand
باشد.
سه گرادیان با فرمول های زیر تعریف می شوند (با فرض یک آرایه 4 بعدی به عنوان operand
و با شاخص بعد ویژگی l
، اندازه دسته ای m
و اندازه های فضایی w
و h
):
\[ \begin{split} c_l&= \frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \left( \nabla y_{ijkl} \frac{x_{ijkl} - \mu_l}{\sigma^2_l+\epsilon} \right) \\\\ d_l&= \frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \nabla y_{ijkl} \\\\ \nabla x_{ijkl} &= \frac{\gamma_{l} }{\sqrt{\sigma^2_{l}+\epsilon} } \left( \nabla y_{ijkl} - d_l - c_l (x_{ijkl} - \mu_{l}) \right) \\\\ \nabla \gamma_l &= \sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \left( \nabla y_{ijkl} \frac{x_{ijkl} - \mu_l}{\sqrt{\sigma^2_{l}+\epsilon} } \right) \\\\\ \nabla \beta_l &= \sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h \nabla y_{ijkl} \end{split} \]
mean
و variance
ورودی ها ارزش گشتاورها را در ابعاد دسته ای و فضایی نشان می دهد.
نوع خروجی سه دسته است:
خروجی ها | تایپ کنید | مفاهیم |
---|---|---|
grad_operand | XlaOp | گرادیان با توجه به operand ورودی (\( \nabla x\)) |
grad_scale | XlaOp | گرادیان با توجه به scale ورودی (\( \nabla \gamma\)) |
grad_offset | XlaOp | گرادیان با توجه به offset ورودی (\( \nabla \beta\)) |
BatchNormInference
همچنین XlaBuilder::BatchNormInference
و مقاله نرمال سازی دسته اصلی را برای توضیح دقیق الگوریتم ببینید.
یک آرایه را در ابعاد دسته ای و فضایی عادی می کند.
BatchNormInference(operand, scale, offset, mean, variance, epsilon, feature_index)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | n آرایه بعدی عادی شود |
scale | XlaOp | آرایه 1 بعدی |
offset | XlaOp | آرایه 1 بعدی |
mean | XlaOp | آرایه 1 بعدی |
variance | XlaOp | آرایه 1 بعدی |
epsilon | float | ارزش اپسیلون |
feature_index | int64 | شاخص به بعد ویژگی در operand |
برای هر ویژگی در بعد ویژگی ( feature_index
شاخص بعد ویژگی در operand
است)، این عملیات میانگین و واریانس را در تمام ابعاد دیگر محاسبه می کند و از میانگین و واریانس برای عادی سازی هر عنصر در operand
استفاده می کند. feature_index
باید یک شاخص معتبر برای بعد ویژگی در operand
باشد.
BatchNormInference
معادل فراخوانی BatchNormTraining
بدون محاسبه mean
و variance
برای هر دسته است. در عوض از mean
ورودی و variance
به عنوان مقادیر تخمینی استفاده می کند. هدف از این عملیات کاهش تأخیر در استنتاج است، از این رو BatchNormInference
نامیده می شود.
خروجی یک آرایه n بعدی و نرمال شده با همان شکل operand
ورودی است.
BatchNormTraining
همچنین XlaBuilder::BatchNormTraining
و the original batch normalization paper
برای توضیح دقیق الگوریتم ببینید.
یک آرایه را در ابعاد دسته ای و فضایی عادی می کند.
BatchNormTraining(operand, scale, offset, epsilon, feature_index)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | n آرایه بعدی باید نرمال شود (x) |
scale | XlaOp | آرایه 1 بعدی (\(\gamma\)) |
offset | XlaOp | آرایه 1 بعدی (\(\beta\)) |
epsilon | float | مقدار اپسیلون (\(\epsilon\)) |
feature_index | int64 | شاخص به بعد ویژگی در operand |
برای هر ویژگی در بعد ویژگی ( feature_index
شاخص بعد ویژگی در operand
است)، این عملیات میانگین و واریانس را در تمام ابعاد دیگر محاسبه می کند و از میانگین و واریانس برای عادی سازی هر عنصر در operand
استفاده می کند. feature_index
باید یک شاخص معتبر برای بعد ویژگی در operand
باشد.
الگوریتم برای هر دسته در operand
\(x\) که شامل m
عناصر با w
و h
به اندازه ابعاد فضایی است (با فرض اینکه operand
یک آرایه 4 بعدی است) به شرح زیر است:
میانگین دسته ای \(\mu_l\) را برای هر ویژگی
l
در بعد ویژگی محاسبه می کند:\(\mu_l=\frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h x_{ijkl}\)محاسبه واریانس دسته ای \(\sigma^2_l\):\(\sigma^2_l=\frac{1}{mwh}\sum_{i=1}^m\sum_{j=1}^w\sum_{k=1}^h (x_{ijkl} - \mu_l)^2\)
عادی، مقیاس و جابجایی:\(y_{ijkl}=\frac{\gamma_l(x_{ijkl}-\mu_l)}{\sqrt[2]{\sigma^2_l+\epsilon} }+\beta_l\)
مقدار اپسیلون، معمولاً یک عدد کوچک، برای جلوگیری از خطاهای تقسیم بر صفر اضافه می شود.
نوع خروجی سه تایی XlaOp
است:
خروجی ها | تایپ کنید | مفاهیم |
---|---|---|
output | XlaOp | n آرایه بعدی با همان شکل operand ورودی (y) |
batch_mean | XlaOp | آرایه 1 بعدی (\(\mu\)) |
batch_var | XlaOp | آرایه 1 بعدی (\(\sigma^2\)) |
batch_mean
و batch_var
گشتاورهایی هستند که در ابعاد دسته ای و فضایی با استفاده از فرمول های بالا محاسبه می شوند.
BitcastConvertType
XlaBuilder::BitcastConvertType
را نیز ببینید.
مشابه یک tf.bitcast
در TensorFlow، عملیات بیتکست عنصری را از شکل داده به شکل هدف انجام می دهد. اندازه ورودی و خروجی باید مطابقت داشته باشند: به عنوان مثال، عناصر s32
از طریق روال بیتکست به عناصر f32
تبدیل میشوند و یک عنصر s32
به چهار عنصر s8
تبدیل میشود. بیتکست بهعنوان یک بازیگر سطح پایین پیادهسازی میشود، بنابراین ماشینهایی با نمایشهای ممیز شناور متفاوت نتایج متفاوتی خواهند داشت.
BitcastConvertType(operand, new_element_type)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه از نوع T با کم نور D |
new_element_type | PrimitiveType | نوع U |
ابعاد عملوند و شکل هدف باید مطابقت داشته باشند، جدای از آخرین بعد که با نسبت اندازه اولیه قبل و بعد از تبدیل تغییر می کند.
نوع عنصر مبدا و مقصد نباید چند تایی باشد.
تبدیل بیتکست به نوع اولیه با عرض های مختلف
دستور BitcastConvert
HLO از حالتی پشتیبانی می کند که اندازه عنصر خروجی نوع T'
با اندازه عنصر ورودی T
برابر نباشد. از آنجایی که کل عملیات از نظر مفهومی یک بیتکست است و بایتهای زیرین را تغییر نمیدهد، شکل عنصر خروجی باید تغییر کند. برای B = sizeof(T), B' = sizeof(T')
، دو حالت ممکن وجود دارد.
ابتدا، وقتی B > B'
، شکل خروجی یک بعد جزئی جدید به اندازه B/B'
می گیرد. مثلا:
f16[10,2]{1,0} %output = f16[10,2]{1,0} bitcast-convert(f32[10]{0} %input)
این قانون برای اسکالرهای موثر یکسان است:
f16[2]{0} %output = f16[2]{0} bitcast-convert(f32[] %input)
روش دیگر، برای B' > B
، دستورالعمل نیاز دارد که آخرین بعد منطقی شکل ورودی برابر با B'/B
باشد، و این بعد در طول تبدیل حذف میشود:
f32[10]{0} %output = f32[10]{0} bitcast-convert(f16[10,2]{1,0} %input)
توجه داشته باشید که تبدیل بین پهنای بیت های مختلف عنصری نیست.
پخش
XlaBuilder::Broadcast
نیز ببینید.
با کپی کردن داده ها در آرایه، ابعاد را به آرایه اضافه می کند.
Broadcast(operand, broadcast_sizes)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه برای کپی کردن |
broadcast_sizes | ArraySlice<int64> | اندازه های ابعاد جدید |
ابعاد جدید در سمت چپ درج می شوند، یعنی اگر broadcast_sizes
دارای مقادیر {a0, ..., aN}
و شکل عملوند دارای ابعاد {b0, ..., bM}
باشد، شکل خروجی دارای ابعاد {a0, ..., aN, b0, ..., bM}
.
ابعاد جدید به کپی هایی از عملوند ایندکس می شود، به عنوان مثال
output[i0, ..., iN, j0, ..., jM] = operand[j0, ..., jM]
به عنوان مثال، اگر operand
اسکالر f32
با مقدار 2.0f
باشد، و broadcast_sizes
{2, 3}
باشد، نتیجه آرایه ای با شکل f32[2, 3]
خواهد بود و تمام مقادیر در نتیجه 2.0f
خواهد بود.
BroadcastInDim
XlaBuilder::BroadcastInDim
نیز ببینید.
اندازه و رتبه یک آرایه را با کپی کردن داده ها در آرایه گسترش می دهد.
BroadcastInDim(operand, out_dim_size, broadcast_dimensions)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه برای کپی کردن |
out_dim_size | ArraySlice<int64> | اندازه ابعاد شکل هدف |
broadcast_dimensions | ArraySlice<int64> | هر بعد از شکل عملوند مربوط به کدام بعد در شکل هدف است |
مشابه Broadcast، اما امکان افزودن ابعاد در هر نقطه و گسترش ابعاد موجود با اندازه 1 را می دهد.
operand
به شکل توصیف شده توسط out_dim_size
پخش می شود. broadcast_dimensions
ابعاد operand
را به ابعاد شکل هدف نگاشت میکند، یعنی بعد i'م عملوند به بعد broadcast_dimension[i]'مین شکل خروجی نگاشت میشود. ابعاد operand
باید اندازه 1 داشته باشد یا به اندازه ابعاد در شکل خروجی که به آن نگاشت شده اند باشد. ابعاد باقیمانده با ابعاد اندازه 1 پر می شود. پخش با ابعاد منحط سپس در امتداد این ابعاد منحط پخش می شود تا به شکل خروجی برسد. معناشناسی به طور مفصل در صفحه پخش توضیح داده شده است.
زنگ زدن
XlaBuilder::Call
نیز ببینید.
محاسبه ای را با آرگومان های داده شده فراخوانی می کند.
Call(computation, args...)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
computation | XlaComputation | محاسبات نوع T_0, T_1, ..., T_{N-1} -> S با N پارامتر از نوع دلخواه |
args | دنباله ای از N XlaOp s | N آرگومان از نوع دلخواه |
آریتی و انواع args
باید با پارامترهای computation
مطابقت داشته باشد. بدون args
مجاز است.
چولسکی
XlaBuilder::Cholesky
نیز ببینید.
تجزیه Cholesky دسته ای از ماتریس های قطعی مثبت متقارن (Hermitian) را محاسبه می کند.
Cholesky(a, lower)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
a | XlaOp | یک آرایه رتبه > 2 از نوع پیچیده یا ممیز شناور. |
lower | bool | از مثلث بالایی یا پایینی a استفاده کنیم. |
اگر lower
true
باشد، ماتریس های مثلثی پایین l
به گونه ای محاسبه می کند که \( a = l
. l^T \). اگر lower
false
باشد، ماتریسهای مثلثی بالا u
بهگونهای محاسبه میکند که \( a = u^T . u \).
بسته به مقدار lower
، داده های ورودی فقط از مثلث پایین/بالایی a
خوانده می شود. مقادیر از مثلث دیگر نادیده گرفته می شوند. داده های خروجی در همان مثلث برگردانده می شوند. مقادیر در مثلث دیگر با پیاده سازی تعریف شده اند و ممکن است هر چیزی باشند.
اگر رتبه a
بزرگتر از 2 باشد، a
به عنوان دسته ای از ماتریس ها در نظر گرفته می شود، که در آن همه ابعاد به جز 2 جزئی، ابعاد دسته ای هستند.
اگر a
متقارن (Hermitian) مثبت قطعی نباشد، نتیجه اجرا تعریف می شود.
گیره
XlaBuilder::Clamp
نیز ببینید.
یک عملوند را در محدوده بین یک مقدار حداقل و حداکثر قرار می دهد.
Clamp(min, operand, max)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
min | XlaOp | آرایه از نوع T |
operand | XlaOp | آرایه از نوع T |
max | XlaOp | آرایه از نوع T |
با توجه به یک عملوند و مقادیر حداقل و حداکثر، اگر در محدوده بین حداقل و حداکثر باشد، عملوند را برمیگرداند، در غیر این صورت اگر عملوند زیر این محدوده باشد، مقدار حداقل و اگر عملوند بالاتر از این محدوده باشد، مقدار حداکثر را برمیگرداند. یعنی clamp(a, x, b) = min(max(a, x), b)
.
هر سه آرایه باید یک شکل باشند. از طرف دیگر، به عنوان یک شکل محدود پخش ، min
و/یا max
میتواند یک اسکالر از نوع T
باشد.
مثال با min
و max
اسکالر:
let operand: s32[3] = {-1, 5, 9};
let min: s32 = 0;
let max: s32 = 6;
==>
Clamp(min, operand, max) = s32[3]{0, 5, 6};
سقوط - فروپاشی
همچنین XlaBuilder::Collapse
و عملیات tf.reshape
را ببینید.
ابعاد یک آرایه را در یک بعد جمع می کند.
Collapse(operand, dimensions)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه از نوع T |
dimensions | وکتور int64 | به ترتیب، زیر مجموعه های متوالی از ابعاد T. |
Collapse زیر مجموعه داده شده از ابعاد عملوند را با یک بعد منفرد جایگزین می کند. آرگومان های ورودی یک آرایه دلخواه از نوع T و یک بردار ثابت-زمان کامپایل از شاخص های بعد هستند. شاخصهای ابعاد باید یک مرتبه (اعداد ابعاد کم تا زیاد)، زیرمجموعهای متوالی از ابعاد T باشند. بنابراین، {0، 1، 2}، {0، 1}، یا {1، 2} همه مجموعههای ابعاد معتبر هستند، اما {1، 0} یا {0، 2} نیستند. آنها با یک بعد جدید جایگزین می شوند، در همان موقعیت در ترتیب ابعادی که جایگزین می شوند، با اندازه ابعاد جدید برابر با حاصلضرب اندازه های ابعاد اصلی. کمترین عدد بعد در dimensions
، کندترین بعد متغیر (بزرگترین) در لانه حلقه است که این ابعاد را جمع می کند، و بالاترین عدد بعد سریعترین تغییر (فرعی ترین) است. در صورت نیاز به سفارش جمعبندی کلیتر، عملگر tf.reshape
را ببینید.
به عنوان مثال، فرض کنید v آرایه ای از 24 عنصر باشد:
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} };
CollectivePermute
XlaBuilder::CollectivePermute
را نیز ببینید.
CollectivePermute یک عملیات جمعی است که داده های متقاطع را ارسال و دریافت می کند.
CollectivePermute(operand, source_target_pairs)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | n آرایه ورودی بعدی |
source_target_pairs | بردار <int64, int64> | لیستی از جفت های (source_replica_id، target_replica_id). برای هر جفت، عملوند از replica مبدا به replica هدف ارسال می شود. |
توجه داشته باشید که محدودیت های زیر برای source_target_pair
وجود دارد:
- هر دو جفت نباید شناسه ماکت هدف یکسانی داشته باشند، و نباید شناسه ماکت منبع یکسانی داشته باشند.
- اگر یک Replica id در هیچ جفتی هدف نباشد، خروجی آن ماکت یک تانسور است که از 0(s) با همان شکل ورودی تشکیل شده است.
الحاق
XlaBuilder::ConcatInDim
را نیز ببینید.
Concatenate یک آرایه از چند عملوند آرایه می سازد. آرایه با هر یک از عملوندهای آرایه ورودی (که باید هم رتبه یکدیگر باشند) رتبه یکسانی دارد و آرگومان ها را به ترتیبی که مشخص کرده اند، در خود دارد.
Concatenate(operands..., dimension)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operands | دنباله ای از N XlaOp | N آرایه از نوع T با ابعاد [L0, L1, ...]. به N >= 1 نیاز دارد. |
dimension | int64 | مقداری در بازه [0, N) که ابعادی را که باید بین operands الحاق شود نام میبرد. |
به استثنای dimension
همه ابعاد باید یکسان باشند. این به این دلیل است که XLA از آرایههای "راگ" پشتیبانی نمیکند. همچنین توجه داشته باشید که مقادیر رتبه-0 را نمی توان به هم متصل کرد (زیرا نام بعدی که در آن الحاق اتفاق می افتد غیرممکن است).
مثال 1 بعدی:
Concat({ {2, 3}, {4, 5}, {6, 7} }, 0)
>>> {2, 3, 4, 5, 6, 7}
مثال دو بعدی:
let a = {
{1, 2},
{3, 4},
{5, 6},
};
let b = {
{7, 8},
};
Concat({a, b}, 0)
>>> {
{1, 2},
{3, 4},
{5, 6},
{7, 8},
}
نمودار:

مشروط
XlaBuilder::Conditional
نیز ببینید.
Conditional(pred, true_operand, true_computation, false_operand, false_computation)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
pred | XlaOp | اسکالر از نوع PRED |
true_operand | XlaOp | آرگومان نوع \(T_0\) |
true_computation | XlaComputation | محاسبات Xla از نوع \(T_0 \to S\) |
false_operand | XlaOp | آرگومان نوع \(T_1\) |
false_computation | XlaComputation | محاسبات Xla از نوع \(T_1 \to S\) |
اگر pred
true
باشد true_computation
اگر pred
false
باشد false_computation
را اجرا میکند و نتیجه را برمیگرداند.
true_computation
باید در یک آرگومان واحد از نوع \(T_0\) باشد و با true_operand
که باید از همان نوع باشد فراخوانی می شود. false_computation
باید یک آرگومان منفرد از نوع \(T_1\) داشته باشد و با false_operand
که باید از همان نوع باشد فراخوانی می شود. نوع مقدار بازگشتی true_computation
و false_computation
باید یکسان باشد.
توجه داشته باشید که فقط یکی از true_computation
و false_computation
بسته به مقدار pred
اجرا خواهد شد.
Conditional(branch_index, branch_computations, branch_operands)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
branch_index | XlaOp | اسکالر از نوع S32 |
branch_computations | دنباله N XlaComputation | محاسبات Xla از نوع \( T_0 \to S , T_1 \to S , ..., T_{N-1} \to S \) |
branch_operands | دنباله ای از N XlaOp | آرگومان های نوع \( T_0 , T_1 , ..., T_{N-1} \) |
branch_computations[branch_index]
را اجرا می کند و نتیجه را برمی گرداند. اگر branch_index
یک S32
است که < 0 یا >= N است، branch_computations[N-1]
به عنوان شاخه پیشفرض اجرا میشود.
هر branch_computations[b]
باید یک آرگومان واحد از نوع T_b
داشته باشد و با branch_operands[b]
که باید از همان نوع باشند فراخوانی می شود. نوع مقدار بازگشتی هر branch_computations[b]
باید یکسان باشد.
توجه داشته باشید که بسته به مقدار branch_index
فقط یکی از branch_computations
اجرا میشود.
تبدیل (پیچیدگی)
XlaBuilder::Conv
نیز ببینید.
به عنوان ConvWithGeneralPadding، اما بالشتک به صورت مختصر به صورت SAME یا VALID مشخص می شود. همان بالشتک ورودی ( lhs
) را با صفر میپوشاند تا در صورت عدم توجه به گام برداشتن، خروجی همان شکل ورودی را داشته باشد. VALID padding به سادگی به معنای عدم وجود بالشتک است.
ConvWithGeneralPadding (کانولوشن)
XlaBuilder::ConvWithGeneralPadding
را نیز ببینید.
یک پیچیدگی از نوع مورد استفاده در شبکه های عصبی را محاسبه می کند. در اینجا، یک کانولوشن را می توان به عنوان یک پنجره n بعدی در نظر گرفت که در یک ناحیه پایه n بعدی حرکت می کند و یک محاسبه برای هر موقعیت احتمالی پنجره انجام می شود.
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
lhs | XlaOp | رتبه n+2 آرایه ورودی |
rhs | XlaOp | رتبه n+2 آرایه از وزن هسته |
window_strides | ArraySlice<int64> | آرایه دوم از گام های هسته |
padding | ArraySlice< pair<int64, int64>> | آرایه دوم از بالشتک (کم، زیاد). |
lhs_dilation | ArraySlice<int64> | آرایه فاکتور اتساع nd lhs |
rhs_dilation | ArraySlice<int64> | آرایه فاکتور اتساع nd rhs |
feature_group_count | int64 | تعداد گروه های ویژگی |
batch_group_count | int64 | تعداد گروه های دسته ای |
n تعداد ابعاد فضایی باشد. آرگومان lhs
یک آرایه رتبه ای n+2 است که ناحیه پایه را توصیف می کند. این ورودی نامیده می شود، حتی اگر البته rhs نیز یک ورودی است. در یک شبکه عصبی، اینها فعال سازی های ورودی هستند. ابعاد n+2 به ترتیب زیر است:
-
batch
: هر مختصات در این بعد یک ورودی مستقل را نشان می دهد که کانولوشن برای آن انجام می شود. -
z/depth/features
: هر موقعیت (y,x) در ناحیه پایه دارای یک بردار مرتبط با آن است که به این بعد می رود. -
spatial_dims
:n
بعد فضایی را توصیف می کند که ناحیه پایه ای را که پنجره در آن حرکت می کند را مشخص می کند.
آرگومان rhs
یک آرایه رتبه ای n+2 است که فیلتر/هسته/پنجره کانولوشنال را توصیف می کند. ابعاد به این ترتیب است:
-
output-z
: بعدz
خروجی. -
input-z
: اندازه این بعد ضربدرfeature_group_count
باید با اندازهz
در lh برابر باشد. -
spatial_dims
:n
بعد فضایی را توصیف می کند که پنجره دومی را که در ناحیه پایه حرکت می کند، تعریف می کند.
آرگومان window_strides
گام پنجره کانولوشن را در ابعاد فضایی مشخص می کند. به عنوان مثال، اگر گام در اولین بعد فضایی 3 باشد، پنجره را فقط می توان در مختصاتی قرار داد که اولین شاخص فضایی بر 3 بخش پذیر باشد.
آرگومان padding
مقدار padding صفر را که باید در ناحیه پایه اعمال شود را مشخص می کند. مقدار padding می تواند منفی باشد -- قدر مطلق padding منفی تعداد عناصری را که باید قبل از انجام کانولوشن از بعد مشخص شده حذف شوند، نشان می دهد. padding[0]
padding را برای بعد y
و padding[1]
padding را برای بعد x
مشخص می کند. هر جفت دارای بالشتک کم به عنوان عنصر اول و بالشتک بالا به عنوان عنصر دوم است. لایه کم در جهت شاخص های پایین تر اعمال می شود در حالی که بالشتک بالا در جهت شاخص های بالاتر اعمال می شود. به عنوان مثال، اگر padding[1]
(2,3)
باشد، در بعد فضایی دوم 2 صفر در سمت چپ و 3 صفر در سمت راست وجود خواهد داشت. استفاده از padding معادل درج همان مقادیر صفر در ورودی ( lhs
) قبل از انجام کانولوشن است.
آرگومان های lhs_dilation
و rhs_dilation
ضریب اتساع را مشخص می کنند که به ترتیب برای lhs و rhs در هر بعد فضایی اعمال می شود. اگر ضریب اتساع در یک بعد فضایی d باشد، سوراخهای d-1 به طور ضمنی بین هر یک از ورودیهای آن بعد قرار میگیرند و اندازه آرایه را افزایش میدهند. سوراخ ها با یک مقدار no-op پر شده اند که برای کانولوشن به معنای صفر است.
به اتساع rhs پیچیدگی آتروس نیز گفته می شود. برای جزئیات بیشتر، tf.nn.atrous_conv2d
را ببینید. اتساع lhs را کانولوشن انتقالی نیز می گویند. برای جزئیات بیشتر، tf.nn.conv2d_transpose
را ببینید.
آرگومان feature_group_count
(مقدار پیش فرض 1) می تواند برای کانولوشن های گروه بندی شده استفاده شود. feature_group_count
باید مقسومکننده بعد ویژگی ورودی و خروجی باشد. اگر feature_group_count
بزرگتر از 1 باشد، به این معنی است که از نظر مفهومی، بعد ویژگی ورودی و خروجی و بعد ویژگی خروجی rhs
feature_group_count
طور مساوی به گروه های بسیاری تقسیم می شوند، که هر گروه متشکل از یک دنباله متوالی از ویژگی ها است. بعد ویژگی ورودی rhs
باید برابر با بعد ویژگی ورودی lhs
تقسیم بر feature_group_count
باشد (بنابراین از قبل اندازه یک گروه از ویژگی های ورودی را دارد). گروه های i-ام با هم برای محاسبه feature_group_count
بسیاری از کانولوشن های جداگانه استفاده می شوند. نتایج این پیچیدگی ها در بعد ویژگی خروجی به هم متصل می شوند.
برای کانولوشن عمیق، آرگومان feature_group_count
روی بعد ویژگی ورودی تنظیم میشود و فیلتر از [filter_height, filter_width, in_channels, channel_multiplier]
به [filter_height, filter_width, 1, in_channels * channel_multiplier]
تغییر شکل میدهد. برای جزئیات بیشتر، tf.nn.depthwise_conv2d
را ببینید.
آرگومان batch_group_count
(مقدار پیشفرض 1) را میتوان برای فیلترهای گروهبندیشده در حین انتشار پسانداز استفاده کرد. batch_group_count
باید مقسومکنندهای برای اندازه lhs
(ورودی) بعد دسته باشد. اگر batch_group_count
بزرگتر از 1 باشد، به این معنی است که بعد دسته خروجی باید به اندازه input batch / batch_group_count
باشد. batch_group_count
باید مقسومکننده اندازه ویژگی خروجی باشد.
شکل خروجی دارای این ابعاد است، به ترتیب:
-
batch
: اندازه این بعد ضربدرbatch_group_count
باید با اندازه بعدbatch
بر حسب lh برابر باشد. -
z
: اندازهoutput-z
در هسته (rhs
). -
spatial_dims
: یک مقدار برای هر قرار دادن معتبر پنجره کانولوشنال.
شکل بالا نحوه کار فیلد batch_group_count
را نشان می دهد. به طور موثر، ما هر دسته lhs را به گروه های batch_group_count
تقسیم می کنیم و همین کار را برای ویژگی های خروجی انجام می دهیم. سپس، برای هر یک از این گروهها، کانولوشنهای زوجی انجام میدهیم و خروجی را در امتداد بعد ویژگی خروجی به هم متصل میکنیم. معنای عملیاتی همه ابعاد دیگر (ویژگی و فضایی) یکسان باقی می ماند.
محل قرارگیری معتبر پنجره کانولوشن با گامها و اندازه ناحیه پایه پس از بالشتک تعیین میشود.
برای توصیف کاری که یک کانولوشن انجام می دهد، یک کانولوشن 2 بعدی را در نظر بگیرید و چند مختصات batch
ثابت، z
، y
، x
را در خروجی انتخاب کنید. سپس (y,x)
موقعیت یک گوشه از پنجره در ناحیه پایه است (مثلاً گوشه سمت چپ بالا، بسته به نحوه تفسیر ابعاد فضایی). اکنون یک پنجره 2 بعدی داریم که از ناحیه پایه گرفته شده است، جایی که هر نقطه 2 بعدی به یک بردار 1 بعدی مرتبط است، بنابراین یک کادر سه بعدی دریافت می کنیم. از هسته کانولوشنال، از آنجایی که مختصات خروجی z
ثابت کردیم، یک جعبه 3 بعدی نیز داریم. ابعاد دو جعبه یکسان است، بنابراین میتوانیم مجموع حاصلضربهای عنصر را بین دو جعبه بگیریم (شبیه به حاصل ضرب نقطهای). این مقدار خروجی است.
توجه داشته باشید که اگر output-z
به عنوان مثال 5 باشد، هر موقعیت از پنجره 5 مقدار در خروجی به بعد z
خروجی تولید می کند. این مقادیر در قسمتی از هسته کانولوشنال استفاده شده متفاوت است - یک جعبه 3 بعدی جداگانه از مقادیر برای هر مختصات output-z
استفاده می شود. بنابراین می توانید آن را به عنوان 5 پیچ مجزا با یک فیلتر متفاوت برای هر یک از آنها در نظر بگیرید.
در اینجا شبه کد برای یک کانولوشن دو بعدی با بالشتک و گام برداشتن آمده است:
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;
}
ConvertElementType
XlaBuilder::ConvertElementType
نیز ببینید.
شبیه به عنصر static_cast
در C++، عملیات تبدیل عنصری را از شکل داده به شکل هدف انجام می دهد. ابعاد باید مطابقت داشته باشند، و تبدیل از نظر عنصر است. به عنوان مثال، عناصر s32
از طریق یک روال تبدیل s32
به f32
به عناصر f32
تبدیل می شوند.
ConvertElementType(operand, new_element_type)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه از نوع T با کم نور D |
new_element_type | PrimitiveType | نوع U |
ابعاد عملوند و شکل هدف باید مطابقت داشته باشند. نوع عنصر مبدا و مقصد نباید چند تایی باشد.
تبدیلی مانند T=s32
به U=f32
یک روال عادی تبدیل درون به شناور مانند دور به نزدیکترین زوج را انجام می دهد.
let a: s32[3] = {0, 1, 2};
let b: f32[3] = convert(a, f32);
then b == f32[3]{0.0, 1.0, 2.0}
CrossReplicaSum
AllReduce
با محاسبات جمع انجام می دهد.
تماس سفارشی
XlaBuilder::CustomCall
نیز ببینید.
یک تابع ارائه شده توسط کاربر را در یک محاسبات فراخوانی کنید.
CustomCall(target_name, args..., shape)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
target_name | string | نام تابع یک دستورالعمل فراخوانی منتشر می شود که این نام نماد را هدف قرار می دهد. |
args | دنباله ای از N XlaOp s | N آرگومان از نوع دلخواه که به تابع ارسال می شود. |
shape | Shape | شکل خروجی تابع |
امضای تابع بدون توجه به آریت یا نوع آرگ ها یکسان است:
extern "C" void target_name(void* out, void** in);
به عنوان مثال، اگر CustomCall به صورت زیر استفاده شود:
let x = f32[2] {1,2};
let y = f32[2x3] { {10, 20, 30}, {40, 50, 60} };
CustomCall("myfunc", {x, y}, f32[3x3])
در اینجا نمونه ای از پیاده سازی myfunc
آمده است:
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];
// ...
}
عملکرد ارائه شده توسط کاربر نباید دارای عوارض جانبی باشد و اجرای آن باید بدون قدرت باشد.
نقطه
XlaBuilder::Dot
نیز ببینید.
Dot(lhs, rhs)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
lhs | XlaOp | آرایه از نوع T |
rhs | XlaOp | آرایه از نوع T |
معنای دقیق این عملیات به رتبه های عملوندها بستگی دارد:
ورودی | خروجی | مفاهیم |
---|---|---|
بردار [n] بردار dot [n] | اسکالر | محصول نقطه برداری |
ماتریس [mxk] بردار dot [k] | بردار [m] | ضرب ماتریس بردار |
ماتریس [mxk] ماتریس dot [kxn] | ماتریس [mxn] | ضرب ماتریس-ماتریس |
این عملیات مجموع محصولات را در بعد دوم lhs
(یا اولین بعد اگر دارای رتبه 1 باشد) و بعد اول rhs
انجام می دهد. اینها ابعاد «قراردادی» هستند. ابعاد انقباض lhs
و rhs
باید به یک اندازه باشند. در عمل، می توان از آن برای انجام محصولات نقطه ای بین بردارها، ضرب بردار/ماتریس یا ضرب ماتریس/ماتریس استفاده کرد.
DotGeneral
XlaBuilder::DotGeneral
نیز ببینید.
DotGeneral(lhs, rhs, dimension_numbers)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
lhs | XlaOp | آرایه از نوع T |
rhs | XlaOp | آرایه از نوع T |
dimension_numbers | DotDimensionNumbers | قرارداد و شماره ابعاد دسته ای |
به عنوان نقطه، اما اجازه می دهد تا اعداد انقباض و ابعاد دسته ای برای هر دو "lhs" و "rhs" مشخص شود.
فیلدهای DotDimensionNumbers | تایپ کنید | مفاهیم |
---|---|---|
'lhs_contracting_dimensions' | تکرار int64 | اعداد ابعاد قراردادی 'lhs' |
'rhs_contracting_dimensions' | تکرار int64 | اعداد ابعاد انقباضی 'rhs' |
'lhs_batch_dimensions' | تکرار int64 | شماره ابعاد دسته ای 'lhs' |
'rhs_batch_dimensions' | تکرار int64 | شماره ابعاد دسته ای 'rhs' |
DotGeneral مجموع محصولات را بر روی ابعاد انقباضی مشخص شده در "dimension_numbers" انجام می دهد.
اعداد ابعاد انقباض مرتبط از «lhs» و «rhs» لازم نیست یکسان باشند، بلکه باید اندازههای ابعاد یکسانی داشته باشند.
مثال با اعداد ابعاد قراردادی:
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} }
اعداد ابعاد دستهای مرتبط از «lhs» و «rhs» باید ابعاد یکسانی داشته باشند.
مثال با اعداد ابعاد دسته ای (متریس اندازه دسته 2، 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} } }
ورودی | خروجی | مفاهیم |
---|---|---|
[b0، m، k] dot [b0، k، n] | [b0، m، n] | دسته ای متمول |
[b0، b1، m، k] dot [b0، b1، k، n] | [b0، b1، m، n] | دسته ای متمول |
نتیجه می شود که عدد بعد حاصل با بعد دسته ای شروع می شود، سپس بعد غیر انقباضی/غیر دسته ای "lhs" و در نهایت بعد غیر انقباضی/غیر دسته ای "rhs".
DynamicSlice
XlaBuilder::DynamicSlice
نیز ببینید.
DynamicSlice یک آرایه فرعی را از آرایه ورودی در start_indices
پویا استخراج می کند. اندازه برش در هر بعد در size_indices
ارسال می شود که نقطه پایان فواصل برش انحصاری در هر بعد را مشخص می کند: [شروع، شروع + اندازه). شکل start_indices
باید رتبه == 1 باشد، با اندازه ابعاد برابر با رتبه operand
.
DynamicSlice(operand, start_indices, size_indices)
استدلال ها | تایپ کنید | مفاهیم |
---|---|---|
operand | XlaOp | آرایه بعدی N از نوع T |
start_indices | دنباله ای از N XlaOp | فهرست N اعداد صحیح اسکالر حاوی شاخص های آغازین برش برای هر بعد. 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:
If
i
is present inbatch_dims
(ie is equal tobatch_dims[k]
for somek
) then we pick the corresponding dimension bounds out ofstart_indices.shape
, skippingindex_vector_dim
(ie pickstart_indices.shape.dims
[k
] ifk
<index_vector_dim
andstart_indices.shape.dims
[k
+1
] otherwise).If
i
is present inoffset_dims
(ie equal tooffset_dims
[k
] for somek
) then we pick the corresponding bound out ofslice_sizes
after accounting forcollapsed_slice_dims
(ie we pickadjusted_slice_sizes
[k
] whereadjusted_slice_sizes
isslice_sizes
with the bounds at indicescollapsed_slice_dims
removed).
Formally, the operand index In
corresponding to a given output index Out
is calculated as follows:
Let
G
= {Out
[k
] fork
inbatch_dims
}. UseG
to slice out a vectorS
such thatS
[i
] =start_indices
[Combine(G
,i
)] where Combine(A, b) inserts b at positionindex_vector_dim
into A. Note that this is well defined even ifG
is empty -- ifG
is empty thenS
=start_indices
.Create a starting index,
S
in
, intooperand
usingS
by scatteringS
usingstart_index_map
. More precisely:S
in
[start_index_map
[k
]] =S
[k
] ifk
<start_index_map.size
.S
in
[_
] =0
otherwise.
Create an index
O
in
intooperand
by scattering the indices at the offset dimensions inOut
according to thecollapsed_slice_dims
set. More precisely:O
in
[remapped_offset_dims
(k
)] =Out
[offset_dims
[k
]] ifk
<offset_dims.size
(remapped_offset_dims
is defined below).O
in
[_
] =0
otherwise.
In
isO
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 { 0
→ 1
, 1
→ 3
, 2
→ 4
, 3
→ 5
}.
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 fromstart_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 theoperand
.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 indexOut
.
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:
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 containingG
0
,G
1
in the last example) are defined to be the output dimensions that are not offset dimensions.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 of1
. Since they have a slice size of1
the only valid index for them is0
and eliding them does not introduce ambiguity.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)
isT
. - If
N > 1
,Collate(T_0, ..., T_{N-1})
is a tuple ofN
elements of typeT
.
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 usingReplicaId
). 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 replicas0
and2
, and1
and3
and then scatters the result. -
shard_count
is the size of each replica group. We need this in cases wherereplica_groups
are empty. Ifreplica_groups
is not empty,shard_count
must be equal to the size of each replica group. -
channel_id
is used for cross-module communication: onlyreduce-scatter
operations with the samechannel_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)
isT
. - If
N > 1
,Collate(T_0, ..., T_{N-1})
is a tuple ofN
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
:
rng_default
: Backend specific algorithm with backend specific shape requirements.rng_three_fry
: ThreeFry counter-based PRNG algorithm. Theinitial_state
shape isu64[2]
with arbitrary values. Salmon et al. SC 2011. Parallel random numbers: as easy as 1, 2, 3.rng_philox
: Philox algorithm to generate random numbers in parallel. Theinitial_state
shape isu64[3]
with arbitrary values. Salmon et al. SC 2011. Parallel random numbers: as easy as 1, 2, 3.
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)
isT
. - If
N > 1
,Collate(T_0, ..., T_N)
is a tuple ofN
elements of typeT
.
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 rankupdate_window_dims.size + scatter_indices.rank - 1
.Bounds of dimension
i
in eachupdates
array must conform to the following:- If
i
is present inupdate_window_dims
(ie equal toupdate_window_dims
[k
] for somek
), then the bound of dimensioni
inupdates
must not exceed the corresponding bound ofoperand
after accounting for theinserted_window_dims
(ieadjusted_window_bounds
[k
], whereadjusted_window_bounds
contains the bounds ofoperand
with the bounds at indicesinserted_window_dims
removed). - If
i
is present inupdate_scatter_dims
(ie equal toupdate_scatter_dims
[k
] for somek
), then the bound of dimensioni
inupdates
must be equal to the corresponding bound ofscatter_indices
, skippingindex_vector_dim
(iescatter_indices.shape.dims
[k
], ifk
<index_vector_dim
andscatter_indices.shape.dims
[k+1
] otherwise).
- If
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 ofupdate_window_dims.size
andinserted_window_dims.size
.scatter_dims_to_operand_dims.size
must be equal toscatter_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:
- Let
G
= {U
[k
] fork
inupdate_scatter_dims
}. UseG
to look up an index vectorS
in thescatter_indices
array such thatS
[i
] =scatter_indices
[Combine(G
,i
)] where Combine(A, b) inserts b at positionsindex_vector_dim
into A. - Create an index
S
in
intooperand
usingS
by scatteringS
using thescatter_dims_to_operand_dims
map. More formally:-
S
in
[scatter_dims_to_operand_dims
[k
]] =S
[k
] ifk
<scatter_dims_to_operand_dims.size
. -
S
in
[_
] =0
otherwise.
-
- Create an index
W
in
into eachoperands
array by scattering the indices atupdate_window_dims
inU
according toinserted_window_dims
. More formally:-
W
in
[window_dims_to_operand_dims
(k
)] =U
[k
] ifk
is inupdate_window_dims
, wherewindow_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, ifupdate_window_dims.size
is4
,operand.rank
is6
, andinserted_window_dims
is {0
,2
} thenwindow_dims_to_operand_dims
is {0
→1
,1
→3
,2
→4
,3
→5
}). -
W
in
[_
] =0
otherwise.
-
-
I
isW
in
+S
in
where + is element-wise addition.
In summary, the scatter operation can be defined as follows.
- Initialize
output
withoperands
, ie for all indicesJ
, for all indicesO
in theoperands
[J
] array:
output
[J
][O
] =operands
[J
][O
] - For every index
U
in theupdates
[J
] array and the corresponding indexO
in theoperand
[J
] array, ifO
is a valid index foroutput
:
(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:
- Current value at the selected index in the output array
- 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 beforeSend
-
Send
happens beforeRecvDone
-
Recv
happens beforeRecvDone
-
Send
happens beforeSendDone
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
withi < j
that eithercomparator(value[i], value[j]) = comparator(value[j], value[i]) = false
orcomparator(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 of1
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
andj
need to be swapped, the comparator is called with2 * n
scalar parameters, where parameter2 * k
corresponds to the value at positioni
from thek-th
operand, and parameter2 * k + 1
corresponds to the value at positionj
from thek-th
operand. Usually, the comparator would thus compare parameters2 * k
and2 * 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 thei-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 typeT
, which is the result from the last execution of thebody
. - 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};
}
