hipDNN backend attribute name C API

2026-03-31

15 min read time

Applies to Linux and Windows

Attribute identifiers for hipDNN backend descriptors.

This file defines the attribute names used with hipdnnBackendSetAttribute() and hipdnnBackendGetAttribute() to configure and query backend descriptors.

Attributes are organized by descriptor type, with each group having a distinct numeric range for identification.

Enums

enum hipdnnBackendAttributeName_t

Identifiers for backend descriptor attributes.

These constants are used with hipdnnBackendSetAttribute() and hipdnnBackendGetAttribute() to specify which attribute to access.

Attribute ranges by descriptor type:

• 100-199: Engine heuristic attributes

• 200-299: Engine configuration attributes

• 300-399: Execution plan attributes

• 400-499: Intermediate info attributes

• 500-599: Knob choice attributes

• 600-699: Operation graph attributes

• 700-799: Variant pack attributes

• 800-899: Layout info attributes

• 900-999: Knob info attributes

• 1000-1099: Engine attributes

• 1100-1199: Kernel cache attributes

• 1200-1299: Device properties attributes

• 1300-1399: Tensor attributes

• 1400-1499: Convolution forward operation attributes

• 1500-1599: Shared convolution descriptor attributes

• 1600-1699: Convolution backward filter operation attributes

• 1700-1799: Convolution backward data operation attributes

• 1800-1899: Batchnorm inference operation attributes

• 1900-1999: Batchnorm inference variance ext operation attributes

• 2000-2099: Batchnorm backward ext operation attributes

• 2100-2199: Shared batchnorm backward ext attributes

• 2200-2299: Pointwise operation attributes

• 2300-2399: Shared pointwise descriptor attributes

• 2400-2499: RMSNorm operation attributes

• 2500-2599: Matmul operation attributes

• 2600-2699: SDPA forward propagation operation attributes

• 2700-2799: Layernorm operation attributes

• 2800-2899: Block scale quantize operation attributes

• 2807-2899: Block scale dequantize operation attributes

• 2900-2913: Batchnorm training forward operation attributes

• 3000-3099: Custom op operation attributes

• 3100-3199: SDPA backward propagation operation attributes

• 60000+: Extension attributes

Values:

enumerator HIPDNN_ATTR_ENGINEHEUR_MODE

Heuristic search mode (hipdnnBackendHeuristicMode_t)

enumerator HIPDNN_ATTR_ENGINEHEUR_OPERATION_GRAPH

Operation graph to find engines for (backend descriptor)

enumerator HIPDNN_ATTR_ENGINEHEUR_RESULTS

Results from heuristic search (array of engine configs)

enumerator HIPDNN_ATTR_ENGINEHEUR_SM_COUNT_TARGET

Target SM count for heuristic evaluation.

enumerator HIPDNN_ATTR_ENGINEHEUR_DEVICEPROP

Device properties for heuristic evaluation.

enumerator HIPDNN_ATTR_ENGINEHEUR_FIND_FIRST_EXT

Find first mode: stop after finding any applicable engine (bool, extension)

enumerator HIPDNN_ATTR_ENGINECFG_ENGINE

The engine this configuration is for (backend descriptor)

enumerator HIPDNN_ATTR_ENGINECFG_INTERMEDIATE_INFO

Intermediate tensor information.

enumerator HIPDNN_ATTR_ENGINECFG_KNOB_CHOICES

Knob choices for this configuration.

enumerator HIPDNN_ATTR_ENGINECFG_WORKSPACE_SIZE

Required workspace size in bytes.

enumerator HIPDNN_ATTR_EXECUTION_PLAN_HANDLE

Associated handle (deprecated, do not use)

enumerator HIPDNN_ATTR_EXECUTION_PLAN_ENGINE_CONFIG

Engine configuration for this plan (backend descriptor)

enumerator HIPDNN_ATTR_EXECUTION_PLAN_WORKSPACE_SIZE

Required workspace size in bytes (int64_t)

enumerator HIPDNN_ATTR_EXECUTION_PLAN_COMPUTED_INTERMEDIATE_UIDS

UIDs of computed intermediate tensors.

enumerator HIPDNN_ATTR_EXECUTION_PLAN_RUN_ONLY_INTERMEDIATE_UIDS

UIDs of run-only intermediate tensors.

enumerator HIPDNN_ATTR_EXECUTION_PLAN_JSON_REPRESENTATION

JSON representation of the execution plan.

enumerator HIPDNN_ATTR_EXECUTION_PLAN_KERNEL_CACHE

Kernel cache for this plan.

enumerator HIPDNN_ATTR_EXECUTION_PLAN_DEVICEPROP

Device properties for this plan.

enumerator HIPDNN_ATTR_INTERMEDIATE_INFO_UNIQUE_ID

Unique ID of the intermediate tensor.

enumerator HIPDNN_ATTR_INTERMEDIATE_INFO_SIZE

Size of the intermediate tensor in bytes.

enumerator HIPDNN_ATTR_INTERMEDIATE_INFO_DEPENDENT_DATA_UIDS

UIDs of dependent input data.

enumerator HIPDNN_ATTR_INTERMEDIATE_INFO_DEPENDENT_ATTRIBUTES

Dependent attribute identifiers.

enumerator HIPDNN_ATTR_KNOB_CHOICE_KNOB_TYPE_EXT

Type of the knob.

enumerator HIPDNN_ATTR_KNOB_CHOICE_KNOB_VALUE_EXT

Selected value for the knob.

enumerator HIPDNN_ATTR_OPERATIONGRAPH_HANDLE

hipDNN handle for this graph

enumerator HIPDNN_ATTR_OPERATIONGRAPH_OPS

Array of operations in the graph.

enumerator HIPDNN_ATTR_OPERATIONGRAPH_ENGINE_GLOBAL_COUNT

Total number of engines available globally.

enumerator HIPDNN_ATTR_OPERATIONGRAPH_IS_DYNAMIC_SHAPE_ENABLED

Whether dynamic shapes are enabled for this graph.

enumerator HIPDNN_ATTR_OPERATIONGRAPH_COMPUTE_DATA_TYPE_EXT

Compute data type for the operation graph (hipdnnDataType_t, extension)

enumerator HIPDNN_ATTR_OPERATIONGRAPH_INTERMEDIATE_DATA_TYPE_EXT

Intermediate data type for the operation graph (hipdnnDataType_t, extension)

enumerator HIPDNN_ATTR_OPERATIONGRAPH_IO_DATA_TYPE_EXT

I/O data type for the operation graph (hipdnnDataType_t, extension)

enumerator HIPDNN_ATTR_OPERATIONGRAPH_PREFERRED_ENGINE_ID_EXT

Preferred engine ID for execution plan selection (int64_t, extension)

enumerator HIPDNN_ATTR_OPERATIONGRAPH_NAME_EXT

Name of the operation graph (HIPDNN_TYPE_CHAR, extension)

enumerator HIPDNN_ATTR_VARIANT_PACK_UNIQUE_IDS

Tensor unique IDs for data pointer mapping.

enumerator HIPDNN_ATTR_VARIANT_PACK_DATA_POINTERS

Device pointers to tensor data.

enumerator HIPDNN_ATTR_VARIANT_PACK_INTERMEDIATES

Intermediate tensor data pointers.

enumerator HIPDNN_ATTR_VARIANT_PACK_WORKSPACE

Workspace pointer for execution.

enumerator HIPDNN_ATTR_LAYOUT_INFO_TENSOR_UID

Tensor UID for layout query.

enumerator HIPDNN_ATTR_LAYOUT_INFO_TYPES

Supported layout types.

enumerator HIPDNN_ATTR_KNOB_INFO_TYPE_EXT

Type identifier of the knob.

enumerator HIPDNN_ATTR_KNOB_INFO_MAXIMUM_VALUE_EXT

Maximum allowed value for the knob.

enumerator HIPDNN_ATTR_KNOB_INFO_MINIMUM_VALUE_EXT

Minimum allowed value for the knob.

enumerator HIPDNN_ATTR_KNOB_INFO_STRIDE_EXT

Step size for valid knob values.

enumerator HIPDNN_ATTR_KNOB_INFO_DESCRIPTION_EXT

Human-readable description of the knob.

enumerator HIPDNN_ATTR_KNOB_INFO_DEFAULT_VALUE_EXT

Default value for the knob (INT64, DOUBLE, or CHAR)

enumerator HIPDNN_ATTR_KNOB_INFO_DEPRECATED_EXT

Whether this knob is deprecated.

enumerator HIPDNN_ATTR_KNOB_INFO_VALID_VALUES_INT_EXT

Explicit list of valid integer values (INT64 array)

enumerator HIPDNN_ATTR_KNOB_INFO_VALID_VALUES_STRING_EXT

Explicit list of valid string values (CHAR, flat null-separated buffer)

enumerator HIPDNN_ATTR_KNOB_INFO_STRING_MAX_LENGTH_EXT

Maximum string length for string knobs (INT32)

enumerator HIPDNN_ATTR_KNOB_INFO_DEFAULT_VALUE_TYPE_EXT

Type discriminator for the default value attribute (read-only, INT64)

Returns the hipdnnBackendAttributeType_t value that should be used when calling getAttribute() for HIPDNN_ATTR_KNOB_INFO_DEFAULT_VALUE_EXT. Possible values: HIPDNN_TYPE_INT64, HIPDNN_TYPE_DOUBLE, HIPDNN_TYPE_CHAR. This eliminates the need to probe multiple types to discover the default value type after reading a KnobDescriptor.

enumerator HIPDNN_ATTR_ENGINE_OPERATION_GRAPH

Operation graph this engine supports.

enumerator HIPDNN_ATTR_ENGINE_GLOBAL_INDEX

Global index of this engine.

enumerator HIPDNN_ATTR_ENGINE_KNOB_INFO

Available knob information for this engine.

enumerator HIPDNN_ATTR_ENGINE_NUMERICAL_NOTE

Numerical behavior notes (precision guarantees)

enumerator HIPDNN_ATTR_ENGINE_LAYOUT_INFO

Layout information for this engine.

enumerator HIPDNN_ATTR_ENGINE_BEHAVIOR_NOTE

Behavioral notes for this engine.

enumerator HIPDNN_ATTR_ENGINE_SM_COUNT_TARGET

Target SM count for this engine.

enumerator HIPDNN_ATTR_ENGINE_DEVICEPROP

Device properties for this engine.

enumerator HIPDNN_ATTR_KERNEL_CACHE_IS_ENGINECFG_KERNEL_CACHED

Whether engine config kernel is cached.

enumerator HIPDNN_ATTR_DEVICEPROP_DEVICE_ID

HIP device ID.

enumerator HIPDNN_ATTR_DEVICEPROP_HANDLE

hipDNN handle associated with device

enumerator HIPDNN_ATTR_DEVICEPROP_JSON_REPRESENTATION

JSON representation of device properties.

enumerator HIPDNN_ATTR_TENSOR_UNIQUE_ID

Unique ID for this tensor.

enumerator HIPDNN_ATTR_TENSOR_NAME_EXT

Tensor name (extension)

enumerator HIPDNN_ATTR_TENSOR_DATA_TYPE

Data type of tensor elements (hipdnnDataType_t)

enumerator HIPDNN_ATTR_TENSOR_DIMENSIONS

Tensor dimensions.

enumerator HIPDNN_ATTR_TENSOR_STRIDES

Tensor strides.

enumerator HIPDNN_ATTR_TENSOR_IS_VIRTUAL

Whether this tensor is virtual.

enumerator HIPDNN_ATTR_TENSOR_VALUE_EXT

Pass-by-value tensor data (extension)

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_FORWARD_W

Weight tensor for forward convolution.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_FORWARD_X

Input tensor for forward convolution.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_FORWARD_Y

Output tensor for forward convolution.

enumerator HIPDNN_ATTR_CONVOLUTION_COMP_TYPE

Compute data type for convolution.

enumerator HIPDNN_ATTR_CONVOLUTION_CONV_MODE

Convolution mode (e.g., cross-correlation)

enumerator HIPDNN_ATTR_CONVOLUTION_DILATIONS

Dilation values for each spatial dimension.

enumerator HIPDNN_ATTR_CONVOLUTION_FILTER_STRIDES

Filter stride values for each spatial dimension.

enumerator HIPDNN_ATTR_CONVOLUTION_POST_PADDINGS

Post-padding values for each spatial dimension.

enumerator HIPDNN_ATTR_CONVOLUTION_PRE_PADDINGS

Pre-padding values for each spatial dimension.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_BACKWARD_FILTER_X

Input tensor for backward filter convolution.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_BACKWARD_FILTER_DY

Output gradient tensor for backward filter convolution.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_BACKWARD_FILTER_DW

Weight gradient tensor for backward filter convolution.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_BACKWARD_DY

Output gradient tensor for backward data convolution.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_BACKWARD_W

Weight tensor for backward data convolution.

enumerator HIPDNN_ATTR_OPERATION_CONVOLUTION_BACKWARD_DX

Input gradient tensor for backward data convolution.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_X_EXT

Input tensor for batchnorm inference.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_MEAN_EXT

Mean tensor for batchnorm inference.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_INV_VARIANCE_EXT

Inverse variance tensor for batchnorm inference.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_SCALE_EXT

Scale tensor for batchnorm inference.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_BIAS_EXT

Bias tensor for batchnorm inference.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_Y_EXT

Output tensor for batchnorm inference.

enumerator HIPDNN_ATTR_BATCHNORM_INF_COMP_TYPE_EXT

Compute data type for batchnorm inference.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_VARIANCE_X_EXT

Input tensor for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_VARIANCE_MEAN_EXT

Mean tensor for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_VARIANCE_VARIANCE_EXT

Variance tensor for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_VARIANCE_SCALE_EXT

Scale tensor for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_VARIANCE_BIAS_EXT

Bias tensor for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_VARIANCE_Y_EXT

Output tensor for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INFERENCE_VARIANCE_EPSILON_EXT

Epsilon tensor for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_BATCHNORM_INF_VAR_COMP_TYPE_EXT

Compute data type for batchnorm inference variance ext.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_DY_EXT

Gradient input tensor (dy) for batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_X_EXT

Input tensor (x) for batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_SCALE_EXT

Scale tensor for batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_DX_EXT

Gradient output tensor (dx) for batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_DSCALE_EXT

Scale gradient tensor (dscale) for batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_DBIAS_EXT

Bias gradient tensor (dbias) for batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_MEAN_EXT

Saved mean tensor from forward pass.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_INV_VARIANCE_EXT

Saved inverse variance tensor from forward pass.

enumerator HIPDNN_ATTR_BATCHNORM_BACKWARD_COMP_TYPE_EXT

Compute data type for batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BACKWARD_PEER_STATS_EXT

Peer statistics tensor array for multi-GPU batchnorm backward.

enumerator HIPDNN_ATTR_OPERATION_POINTWISE_IN_0_EXT

Primary input tensor for pointwise operation.

enumerator HIPDNN_ATTR_OPERATION_POINTWISE_OUT_0_EXT

Output tensor for pointwise operation.

enumerator HIPDNN_ATTR_OPERATION_POINTWISE_IN_1_EXT

Secondary input tensor for pointwise operation (binary/ternary)

enumerator HIPDNN_ATTR_OPERATION_POINTWISE_IN_2_EXT

Tertiary input tensor for pointwise operation (ternary)

enumerator HIPDNN_ATTR_POINTWISE_MODE

Pointwise operation mode.

enumerator HIPDNN_ATTR_POINTWISE_RELU_LOWER_CLIP

Lower clip value for ReLU activation.

enumerator HIPDNN_ATTR_POINTWISE_RELU_UPPER_CLIP

Upper clip value for ReLU activation.

enumerator HIPDNN_ATTR_POINTWISE_RELU_LOWER_CLIP_SLOPE

Lower clip slope for leaky ReLU activation.

enumerator HIPDNN_ATTR_POINTWISE_SWISH_BETA

Beta parameter for Swish activation.

enumerator HIPDNN_ATTR_POINTWISE_ELU_ALPHA

Alpha parameter for ELU activation.

enumerator HIPDNN_ATTR_POINTWISE_SOFTPLUS_BETA

Beta parameter for Softplus activation.

enumerator HIPDNN_ATTR_POINTWISE_MATH_PREC

Compute data type for pointwise operation.

enumerator HIPDNN_ATTR_POINTWISE_AXIS

Axis index for pointwise operation.

enumerator HIPDNN_ATTR_OPERATION_RMSNORM_X_EXT

Input tensor (X) for rmsnorm.

enumerator HIPDNN_ATTR_OPERATION_RMSNORM_SCALE_EXT

Scale tensor for rmsnorm.

enumerator HIPDNN_ATTR_OPERATION_RMSNORM_EPSILON_EXT

Epsilon tensor for rmsnorm.

enumerator HIPDNN_ATTR_OPERATION_RMSNORM_Y_EXT

Output tensor (Y) for rmsnorm.

enumerator HIPDNN_ATTR_OPERATION_RMSNORM_BIAS_EXT

Bias tensor for rmsnorm (optional)

enumerator HIPDNN_ATTR_OPERATION_RMSNORM_INV_RMS_EXT

Inverse RMS tensor for rmsnorm (optional, training only)

enumerator HIPDNN_ATTR_OPERATION_RMSNORM_FWD_PHASE_EXT

Forward phase for rmsnorm (TRAINING or INFERENCE)

enumerator HIPDNN_ATTR_RMSNORM_MATH_PREC_EXT

Compute data type for rmsnorm. Note: intentionally omits OPERATION_ prefix to match the BatchNorm inference convention (HIPDNN_ATTR_BATCHNORM_INF_COMP_TYPE_EXT).

enumerator HIPDNN_ATTR_OPERATION_MATMUL_A_EXT

Left input matrix tensor (A) for matmul.

enumerator HIPDNN_ATTR_OPERATION_MATMUL_B_EXT

Right input matrix tensor (B) for matmul.

enumerator HIPDNN_ATTR_OPERATION_MATMUL_C_EXT

Output matrix tensor (C) for matmul.

enumerator HIPDNN_ATTR_MATMUL_MATH_PREC_EXT

Compute data type for matmul.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_Q_EXT

Q (query) tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_K_EXT

K (key) tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_V_EXT

V (value) tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_O_EXT

O (output) tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_ATTN_MASK_EXT

Attention mask tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SCALE_EXT

Scale tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SEQ_LEN_Q_EXT

Sequence length Q tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SEQ_LEN_KV_EXT

Sequence length KV tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SEED_EXT

Seed tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_OFFSET_EXT

Offset tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_DROPOUT_MASK_EXT

Dropout mask tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_DROPOUT_SCALE_EXT

Dropout scale tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_PAGE_TABLE_K_EXT

Page table K tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_PAGE_TABLE_V_EXT

Page table V tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_BLOCK_MASK_EXT

Block mask tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SINK_TOKEN_EXT

Sink token tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_DESCALE_Q_EXT

Descale Q tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_DESCALE_K_EXT

Descale K tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_DESCALE_V_EXT

Descale V tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_DESCALE_S_EXT

Descale S tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SCALE_S_EXT

Scale S tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SCALE_O_EXT

Scale O tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_STATS_EXT

Stats output tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_MAX_EXT

Max output tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_SUM_EXP_EXT

Sum exp output tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_RNG_DUMP_EXT

RNG dump output tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_AMAX_S_EXT

Amax S output tensor for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_FPROP_AMAX_O_EXT

Amax O output tensor for SDPA forward.

enumerator HIPDNN_ATTR_SDPA_FPROP_GENERATE_STATS_EXT

Whether to generate statistics (bool)

enumerator HIPDNN_ATTR_SDPA_FPROP_ALIBI_MASK_EXT

Whether to use ALiBi mask (bool)

enumerator HIPDNN_ATTR_SDPA_FPROP_PADDING_MASK_EXT

Whether to use padding mask (bool)

enumerator HIPDNN_ATTR_SDPA_FPROP_CAUSAL_MASK_EXT

Whether to use causal mask (bool, deprecated)

enumerator HIPDNN_ATTR_SDPA_FPROP_CAUSAL_MASK_BOTTOM_RIGHT_EXT

Whether to use causal mask bottom-right (bool, deprecated)

enumerator HIPDNN_ATTR_SDPA_FPROP_DROPOUT_PROBABILITY_EXT

Dropout probability (float)

enumerator HIPDNN_ATTR_SDPA_FPROP_ATTN_SCALE_VALUE_EXT

Attention scale value (float)

enumerator HIPDNN_ATTR_SDPA_FPROP_LEFT_BOUND_EXT

Left bound for sliding window (int64)

enumerator HIPDNN_ATTR_SDPA_FPROP_RIGHT_BOUND_EXT

Right bound for sliding window (int64)

enumerator HIPDNN_ATTR_SDPA_FPROP_MAX_SEQ_LEN_KV_EXT

Maximum sequence length KV (int32_t)

enumerator HIPDNN_ATTR_SDPA_FPROP_DIAGONAL_ALIGNMENT_EXT

Diagonal alignment mode (hipdnnDiagonalAlignment_t)

enumerator HIPDNN_ATTR_SDPA_FPROP_MMA_CORE_MODE_EXT

MMA core mode (hipdnnDataType_t)

enumerator HIPDNN_ATTR_SDPA_FPROP_IMPLEMENTATION_EXT

Attention implementation mode (hipdnnAttentionImplementation_t)

enumerator HIPDNN_ATTR_SDPA_FPROP_MATH_PREC_EXT

Compute data type for SDPA forward.

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_X_EXT

Input tensor for layernorm.

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_SCALE_EXT

Scale tensor for layernorm.

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_BIAS_EXT

Bias tensor for layernorm.

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_EPSILON_EXT

Epsilon tensor for layernorm.

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_Y_EXT

Output tensor for layernorm.

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_MEAN_EXT

Mean output tensor for layernorm (optional)

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_INV_VARIANCE_EXT

Inverse variance output tensor for layernorm (optional)

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_FWD_PHASE_EXT

Forward phase for layernorm (TRAINING or INFERENCE)

enumerator HIPDNN_ATTR_LAYERNORM_MATH_PREC_EXT

Math precision (compute data type) for layernorm.

enumerator HIPDNN_ATTR_OPERATION_LAYERNORM_NORMALIZED_DIM_COUNT_EXT

Number of normalized dimensions for layernorm.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_QUANTIZE_X_EXT

Input tensor (X) for block scale quantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_QUANTIZE_Y_EXT

Output tensor (Y) for block scale quantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_QUANTIZE_SCALE_EXT

Scale tensor for block scale quantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_QUANTIZE_BLOCK_SIZE_EXT

Block size parameter for block scale quantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_QUANTIZE_AXIS_EXT

Axis parameter for block scale quantize (optional)

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_QUANTIZE_TRANSPOSE_EXT

Transpose parameter for block scale quantize.

enumerator HIPDNN_ATTR_BLOCK_SCALE_QUANTIZE_MATH_PREC_EXT

Math precision for block scale quantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_DEQUANTIZE_X_EXT

Input tensor for block scale dequantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_DEQUANTIZE_SCALE_EXT

Scale tensor for block scale dequantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_DEQUANTIZE_Y_EXT

Output tensor for block scale dequantize.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_DEQUANTIZE_BLOCK_SIZE_EXT

Block sizes for each dimension.

enumerator HIPDNN_ATTR_OPERATION_BLOCK_SCALE_DEQUANTIZE_IS_NEGATIVE_SCALE_EXT

Whether scale is negative.

enumerator HIPDNN_ATTR_BLOCK_SCALE_DEQUANTIZE_MATH_PREC_EXT

Math precision for block scale dequantize.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_X_EXT

Input tensor (X) for batchnorm training forward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_SCALE_EXT

Scale tensor for batchnorm training forward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_BIAS_EXT

Bias tensor for batchnorm training forward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_EPSILON_EXT

Epsilon tensor for batchnorm training forward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_Y_EXT

Output tensor (Y) for batchnorm training forward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_PREV_RUNNING_MEAN_EXT

Previous running mean tensor (optional)

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_PREV_RUNNING_VARIANCE_EXT

Previous running variance tensor (optional)

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_MOMENTUM_EXT

Momentum tensor (optional, required if running stats provided)

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_MEAN_EXT

Batch mean output tensor (optional)

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_INV_VARIANCE_EXT

Batch inverse variance output tensor (optional)

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_NEXT_RUNNING_MEAN_EXT

Next running mean output tensor (optional)

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_NEXT_RUNNING_VARIANCE_EXT

Next running variance output tensor (optional)

enumerator HIPDNN_ATTR_BATCHNORM_MATH_PREC_EXT

Math precision (compute data type) for batchnorm training forward.

enumerator HIPDNN_ATTR_OPERATION_BATCHNORM_PEER_STATS_EXT

Peer statistics tensor array for multi-GPU batchnorm training forward.

enumerator HIPDNN_ATTR_OPERATION_CUSTOM_OP_INPUTS_EXT

Input tensor array for custom op.

enumerator HIPDNN_ATTR_OPERATION_CUSTOM_OP_OUTPUTS_EXT

Output tensor array for custom op.

enumerator HIPDNN_ATTR_OPERATION_CUSTOM_OP_ID_EXT

Plugin identifier string for custom op (e.g. “example.rope”)

enumerator HIPDNN_ATTR_OPERATION_CUSTOM_OP_DATA_EXT

Opaque byte payload for custom op.

enumerator HIPDNN_ATTR_CUSTOM_OP_COMP_TYPE_EXT

Compute data type for custom op.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_Q_EXT

Q (query) tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_K_EXT

K (key) tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_V_EXT

V (value) tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_O_EXT

O (output) tensor from forward pass for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DO_EXT

dO (output gradient) tensor for SDPA backward

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_STATS_EXT

Stats tensor from forward pass for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DQ_EXT

dQ (query gradient) output tensor for SDPA backward

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DK_EXT

dK (key gradient) output tensor for SDPA backward

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DV_EXT

dV (value gradient) output tensor for SDPA backward

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_SCALE_EXT

Scale tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_ATTN_MASK_EXT

Attention mask tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_SEQ_LEN_Q_EXT

Sequence length Q tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_SEQ_LEN_KV_EXT

Sequence length KV tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_SEED_EXT

Seed tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_OFFSET_EXT

Offset tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DROPOUT_MASK_EXT

Dropout mask tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DROPOUT_SCALE_EXT

Dropout scale tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DROPOUT_SCALE_INV_EXT

Dropout scale inverse tensor for SDPA backward.

enumerator HIPDNN_ATTR_OPERATION_SDPA_BPROP_DBIAS_EXT

dBias (gradient of attention bias) output tensor for SDPA backward

enumerator HIPDNN_ATTR_SDPA_BPROP_ALIBI_MASK_EXT

Whether to use ALiBi mask (bool)

enumerator HIPDNN_ATTR_SDPA_BPROP_PADDING_MASK_EXT

Whether to use padding mask (bool)

enumerator HIPDNN_ATTR_SDPA_BPROP_CAUSAL_MASK_EXT

Whether to use causal mask (bool, deprecated)

enumerator HIPDNN_ATTR_SDPA_BPROP_CAUSAL_MASK_BOTTOM_RIGHT_EXT

Whether to use causal mask bottom-right (bool, deprecated)

enumerator HIPDNN_ATTR_SDPA_BPROP_DROPOUT_PROBABILITY_EXT

Dropout probability (float)

enumerator HIPDNN_ATTR_SDPA_BPROP_ATTN_SCALE_VALUE_EXT

Attention scale value (float)

enumerator HIPDNN_ATTR_SDPA_BPROP_LEFT_BOUND_EXT

Left bound for sliding window (int64)

enumerator HIPDNN_ATTR_SDPA_BPROP_RIGHT_BOUND_EXT

Right bound for sliding window (int64)

enumerator HIPDNN_ATTR_SDPA_BPROP_DIAGONAL_ALIGNMENT_EXT

Diagonal alignment mode (hipdnnDiagonalAlignment_t)

enumerator HIPDNN_ATTR_SDPA_BPROP_MATH_PREC_EXT

Compute data type for SDPA backward.

enumerator HIPDNN_ATTR_KNOB_INFO_SERIALIZED_VALUE_EXT

Serialized knob info as FlatBuffer (extension)

Used to retrieve knob information in serialized FlatBuffer format. Type: HIPDNN_TYPE_FLATBUFFER_DATA_STRUCT_EXT

enumerator HIPDNN_ATTR_KNOB_CHOICE_SERIALIZED_VALUE_EXT

Serialized knob choice as FlatBuffer (extension)

Used to set knob values in serialized FlatBuffer format. Type: HIPDNN_TYPE_FLATBUFFER_DATA_STRUCT_EXT

enumerator HIPDNN_ATTR_OPERATION_TYPE_EXT

Operation type of an operation descriptor (read-only extension)

Returns the hipdnnOperationType_t of an operation descriptor, enabling type-based dispatch without trial-and-error probing. Type: HIPDNN_TYPE_OPERATION_TYPE_EXT

enumerator HIPDNN_ATTR_OPERATION_NAME_EXT

Name of an operation descriptor (extension)

Gets or sets a human-readable name for an operation node, useful for debugging, logging, and round-tripping through serialized graphs. Type: HIPDNN_TYPE_CHAR