98 inline bool hasTensor(IOperationExecutionContext* context, const uint32_t tensor) {  in hasTensor()  argument
99     return context->getInputBuffer(tensor) != nullptr;  in hasTensor()
106 bool validate(const IOperationValidationContext* context) {  in validate()  argument
107     NN_RET_CHECK_EQ(context->getNumInputs(), kNumInputs);  in validate()
108     NN_RET_CHECK_EQ(context->getNumOutputs(), kNumOutputs);  in validate()
143     NN_RET_CHECK(validateInputTypes(context, inExpectedTypes));  in validate()
152     NN_RET_CHECK(validateOutputTypes(context, outExpectedTypes));  in validate()
154     return validateHalVersion(context, HalVersion::V1_3);  in validate()
157 bool prepare(IOperationExecutionContext* context) {  in prepare()  argument
174         NN_RET_CHECK(!context->isOmittedInput(tensor))  in prepare()
178     const Shape inputShape = context->getInputShape(kInputTensor);  in prepare()
185     const Shape inputToOutputShape = context->getInputShape(kInputToOutputWeightsTensor);  in prepare()
190     const Shape recurrentToOutputShape = context->getInputShape(kRecurrentToOutputWeightsTensor);  in prepare()
195     if (hasTensor(context, kInputToInputWeightsTensor)) {  in prepare()
196         const Shape inputToInputShape = context->getInputShape(kInputToInputWeightsTensor);  in prepare()
202     const Shape inputToForgetShape = context->getInputShape(kInputToForgetWeightsTensor);  in prepare()
206     const Shape inputToCellShape = context->getInputShape(kInputToCellWeightsTensor);  in prepare()
211     if (hasTensor(context, kRecurrentToInputWeightsTensor)) {  in prepare()
212         const Shape recurrentToInputShape = context->getInputShape(kRecurrentToInputWeightsTensor);  in prepare()
218     const Shape recurrentToForgetShape = context->getInputShape(kRecurrentToForgetWeightsTensor);  in prepare()
222     const Shape recurrentToCellShape = context->getInputShape(kRecurrentToCellWeightsTensor);  in prepare()
229     const bool cifgWeightsAllOrNone = (hasTensor(context, kInputToInputWeightsTensor) &&  in prepare()
230                                        hasTensor(context, kRecurrentToInputWeightsTensor)) ||  in prepare()
231                                       (!hasTensor(context, kInputToInputWeightsTensor) &&  in prepare()
232                                        !hasTensor(context, kRecurrentToInputWeightsTensor));  in prepare()
235     if (hasTensor(context, kCellToInputWeightsTensor)) {  in prepare()
236         const Shape cellToInputShape = context->getInputShape(kCellToInputWeightsTensor);  in prepare()
241     if (hasTensor(context, kCellToForgetWeightsTensor)) {  in prepare()
242         const Shape cellToForgetShape = context->getInputShape(kCellToForgetWeightsTensor);  in prepare()
247     if (hasTensor(context, kCellToOutputWeightsTensor)) {  in prepare()
248         const Shape cellToOutputShape = context->getInputShape(kCellToOutputWeightsTensor);  in prepare()
254     const bool cifgUsed = !hasTensor(context, kInputToInputWeightsTensor);  in prepare()
256             ((hasTensor(context, kCellToInputWeightsTensor) || cifgUsed) &&  in prepare()
257              hasTensor(context, kCellToForgetWeightsTensor) &&  in prepare()
258              hasTensor(context, kCellToOutputWeightsTensor)) ||  in prepare()
259             (!hasTensor(context, kCellToInputWeightsTensor) &&  in prepare()
260              !hasTensor(context, kCellToForgetWeightsTensor) &&  in prepare()
261              !hasTensor(context, kCellToOutputWeightsTensor));  in prepare()
265         NN_RET_CHECK(hasTensor(context, kInputGateBiasTensor));  in prepare()
266         const Shape inputGateBiasShape = context->getInputShape(kInputGateBiasTensor);  in prepare()
270         NN_RET_CHECK(!hasTensor(context, kInputGateBiasTensor))  in prepare()
274     const Shape forgetGateBiasShape = context->getInputShape(kForgetGateBiasTensor);  in prepare()
277     const Shape cellGateBiasShape = context->getInputShape(kCellGateBiasTensor);  in prepare()
280     const Shape outputGateBiasShape = context->getInputShape(kOutputGateBiasTensor);  in prepare()
284     if (hasTensor(context, kProjectionWeightsTensor)) {  in prepare()
285         const Shape projectionShape = context->getInputShape(kProjectionWeightsTensor);  in prepare()
291     if (hasTensor(context, kProjectionBiasTensor)) {  in prepare()
292         const Shape projectionBiasShape = context->getInputShape(kProjectionBiasTensor);  in prepare()
297     const Shape outputStateShape = context->getInputShape(kPrevOutputTensor);  in prepare()
301     const Shape cellStateShape = context->getInputShape(kPrevCellStateTensor);  in prepare()
306     if (hasTensor(context, kInputLayerNormTensor)) {  in prepare()
307         const Shape inputLayerNormShape = context->getInputShape(kInputLayerNormTensor);  in prepare()
312     if (hasTensor(context, kForgetLayerNormTensor)) {  in prepare()
313         const Shape forgetLayerNormShape = context->getInputShape(kForgetLayerNormTensor);  in prepare()
318     if (hasTensor(context, kCellLayerNormTensor)) {  in prepare()
319         const Shape cellLayerNormShape = context->getInputShape(kCellLayerNormTensor);  in prepare()
324     if (hasTensor(context, kOutputLayerNormTensor)) {  in prepare()
325         const Shape outputLayerNormShape = context->getInputShape(kOutputLayerNormTensor);  in prepare()
331         NN_RET_CHECK(!hasTensor(context, kInputLayerNormTensor))  in prepare()
333         const bool layerNormWeightsAllOrNoneCifg = (hasTensor(context, kForgetLayerNormTensor) &&  in prepare()
334                                                     hasTensor(context, kCellLayerNormTensor) &&  in prepare()
335                                                     hasTensor(context, kOutputLayerNormTensor)) ||  in prepare()
336                                                    (!hasTensor(context, kForgetLayerNormTensor) &&  in prepare()
337                                                     !hasTensor(context, kCellLayerNormTensor) &&  in prepare()
338                                                     !hasTensor(context, kOutputLayerNormTensor));  in prepare()
341         const bool layerNormWeightsAllOrNone = (hasTensor(context, kInputLayerNormTensor) &&  in prepare()
342                                                 hasTensor(context, kForgetLayerNormTensor) &&  in prepare()
343                                                 hasTensor(context, kCellLayerNormTensor) &&  in prepare()
344                                                 hasTensor(context, kOutputLayerNormTensor)) ||  in prepare()
345                                                (!hasTensor(context, kInputLayerNormTensor) &&  in prepare()
346                                                 !hasTensor(context, kForgetLayerNormTensor) &&  in prepare()
347                                                 !hasTensor(context, kCellLayerNormTensor) &&  in prepare()
348                                                 !hasTensor(context, kOutputLayerNormTensor));  in prepare()
352     const Shape prevOutputShape = context->getInputShape(kPrevOutputTensor);  in prepare()
353     Shape outputShape = context->getOutputShape(kOutputTensor);  in prepare()
356     const Shape prevCellStateShape = context->getInputShape(kPrevCellStateTensor);  in prepare()
357     Shape cellStateOutShape = context->getOutputShape(kCellStateOutTensor);  in prepare()
360     return context->setOutputShape(kOutputStateOutTensor, outputShape) &&  in prepare()
361            context->setOutputShape(kCellStateOutTensor, cellStateOutShape) &&  in prepare()
362            context->setOutputShape(kOutputTensor, outputShape);  in prepare()
365 bool execute(IOperationExecutionContext* context) {  in execute()  argument
367     const Shape inputShape = context->getInputShape(kInputTensor);  in execute()
368     const Shape inputToInputWeightsShape = context->getInputShape(kInputToInputWeightsTensor);  in execute()
370             context->getInputShape(kRecurrentToInputWeightsTensor);  in execute()
371     const Shape cellToInputShape = context->getInputShape(kCellToInputWeightsTensor);  in execute()
372     const Shape inputLayerNormShape = context->getInputShape(kInputLayerNormTensor);  in execute()
373     const Shape inputToForgetWeightsShape = context->getInputShape(kInputToForgetWeightsTensor);  in execute()
375             context->getInputShape(kRecurrentToForgetWeightsTensor);  in execute()
376     const Shape cellToForgetShape = context->getInputShape(kCellToForgetWeightsTensor);  in execute()
377     const Shape forgetLayerNormShape = context->getInputShape(kForgetLayerNormTensor);  in execute()
378     const Shape inputToCellWeightsShape = context->getInputShape(kInputToCellWeightsTensor);  in execute()
379     const Shape recurrentToCellWeightsShape = context->getInputShape(kRecurrentToCellWeightsTensor);  in execute()
380     const Shape cellLayerNormShape = context->getInputShape(kCellLayerNormTensor);  in execute()
381     const Shape inputToOutputWeightsShape = context->getInputShape(kInputToOutputWeightsTensor);  in execute()
383             context->getInputShape(kRecurrentToOutputWeightsTensor);  in execute()
384     const Shape cellToOutputShape = context->getInputShape(kCellToOutputWeightsTensor);  in execute()
385     const Shape outputLayerNormShape = context->getInputShape(kOutputLayerNormTensor);  in execute()
386     const Shape projectionWeightsShape = context->getInputShape(kProjectionWeightsTensor);  in execute()
387     const Shape prevOutputShape = context->getInputShape(kPrevOutputTensor);  in execute()
388     const Shape prevCellStateShape = context->getInputShape(kPrevCellStateTensor);  in execute()
395     const float cellClip = context->getInputValue<float>(kCellClip);  in execute()
396     const float projectionClip = context->getInputValue<float>(kProjectionClip);  in execute()
397     const float inputIntermediateScale = context->getInputValue<float>(kInputIntermediateScale);  in execute()
398     const float forgetIntermediateScale = context->getInputValue<float>(kForgetIntermediateScale);  in execute()
399     const float cellIntermediateScale = context->getInputValue<float>(kCellIntermediateScale);  in execute()
400     const float outputIntermediateScale = context->getInputValue<float>(kOutputIntermediateScale);  in execute()
401     const int8_t hiddenStateZeroPoint = context->getInputValue<int8_t>(kHiddenStateZeroPoint);  in execute()
402     const float hiddenStateScale = context->getInputValue<float>(kHiddenStateScale);  in execute()
405             reinterpret_cast<const int8_t*>(context->getInputBuffer(kInputTensor));  in execute()
408             reinterpret_cast<const int8_t*>(context->getInputBuffer(kInputToInputWeightsTensor));  in execute()
411             context->getInputBuffer(kRecurrentToInputWeightsTensor));  in execute()
413             reinterpret_cast<const int16_t*>(context->getInputBuffer(kCellToInputWeightsTensor));  in execute()
415             reinterpret_cast<const int16_t*>(context->getInputBuffer(kInputLayerNormTensor));  in execute()
417             reinterpret_cast<const int32_t*>(context->getInputBuffer(kInputGateBiasTensor));  in execute()
420             reinterpret_cast<const int8_t*>(context->getInputBuffer(kInputToForgetWeightsTensor));  in execute()
422             context->getInputBuffer(kRecurrentToForgetWeightsTensor));  in execute()
424             reinterpret_cast<const int16_t*>(context->getInputBuffer(kCellToForgetWeightsTensor));  in execute()
426             reinterpret_cast<const int16_t*>(context->getInputBuffer(kForgetLayerNormTensor));  in execute()
428             reinterpret_cast<const int32_t*>(context->getInputBuffer(kForgetGateBiasTensor));  in execute()
431             reinterpret_cast<const int8_t*>(context->getInputBuffer(kInputToCellWeightsTensor));  in execute()
433             reinterpret_cast<const int8_t*>(context->getInputBuffer(kRecurrentToCellWeightsTensor));  in execute()
435             reinterpret_cast<const int16_t*>(context->getInputBuffer(kCellLayerNormTensor));  in execute()
437             reinterpret_cast<const int32_t*>(context->getInputBuffer(kCellGateBiasTensor));  in execute()
440             reinterpret_cast<const int8_t*>(context->getInputBuffer(kInputToOutputWeightsTensor));  in execute()
442             context->getInputBuffer(kRecurrentToOutputWeightsTensor));  in execute()
444             reinterpret_cast<const int16_t*>(context->getInputBuffer(kCellToOutputWeightsTensor));  in execute()
446             reinterpret_cast<const int16_t*>(context->getInputBuffer(kOutputLayerNormTensor));  in execute()
448             reinterpret_cast<const int32_t*>(context->getInputBuffer(kOutputGateBiasTensor));  in execute()
451             reinterpret_cast<const int8_t*>(context->getInputBuffer(kProjectionWeightsTensor));  in execute()
453             reinterpret_cast<const int32_t*>(context->getInputBuffer(kProjectionBiasTensor));  in execute()
456             reinterpret_cast<const int8_t*>(context->getInputBuffer(kPrevOutputTensor));  in execute()
458             reinterpret_cast<const int16_t*>(context->getInputBuffer(kPrevCellStateTensor));  in execute()
461             reinterpret_cast<uint8_t*>(context->getOutputBuffer(kOutputStateOutTensor));  in execute()
463             reinterpret_cast<int16_t*>(context->getOutputBuffer(kCellStateOutTensor));  in execute()
464     int8_t* outputBuffer = reinterpret_cast<int8_t*>(context->getOutputBuffer(kOutputTensor));  in execute()