AOCL-DLP API Documentation

AOCL-DLP API Documentation#

Welcome to the AMD Optimizing CPU Libraries - Deep Learning Primitives (AOCL-DLP) API documentation.

This site provides the API reference only. For user guides, tutorials, installation, and examples, please visit the project Wiki.

API Reference#

Complete API reference for AOCL-DLP (AMD Optimizing CPU Libraries - Deep Learning Primitives).

API Categories

Quick API Lookup#

Core GEMM Operations#

GEMM API Functions#

Function Pattern

Description

aocl_gemm_f32f32f32of32

Float32 precision GEMM

aocl_gemm_bf16bf16f32of32

BFloat16 inputs, float32 output

aocl_gemm_u8s8s32os32

Unsigned/signed 8-bit quantized GEMM

aocl_gemm_s8s8s32os8

Signed 8-bit quantized GEMM

aocl_gemm_f16f16f16of16

IEEE float16 precision GEMM

aocl_gemm_bf16s4f32of32

BFloat16 x int4 mixed precision

aocl_gemm_bf16s8s32os32

BFloat16 x int8 mixed precision

aocl_gemm_f32s8s32os32

Float32 x int8 mixed precision

aocl_gemm_s8s8s32of32_sym_quant

Symmetric quantization GEMM

Batch Operations#

Batch GEMM Functions#

Function Pattern

Description

aocl_batch_gemm_*

Batch processing for multiple matrices

Post-Operations#

Post-Operations Framework#

Type / Structure

Description

dlp_metadata_t

Main metadata structure for configuring post-ops

DLP_POST_OP_TYPE

Post-op types: BIAS, ELTWISE, SCALE, MATRIX_ADD, MATRIX_MUL

DLP_ELT_ALGO_TYPE

Activation functions: RELU, GELU, SWISH, TANH, SIGMOID, etc.

Matrix Utilities#

Matrix Utility Functions#

Function Pattern

Description

aocl_get_reorder_buf_size_*

Get buffer size for matrix reordering

aocl_reorder_*

Reorder matrix for optimal performance

aocl_unreorder_*

Convert reordered matrix back to normal format

Element-wise Operations#

Element-wise Functions#

Function Pattern

Description

aocl_gemm_eltwise_ops_*

Apply element-wise operations to matrices

Utility Functions#

Utility Functions#

Function Pattern

Description

aocl_gemm_gelu_*

GELU activation functions

aocl_gemm_softmax_*

Softmax functions

Library Management#

Library Functions#

Function

Description

dlp_thread_set_num_threads

Configure thread count

dlp_thread_set_ways

Configure parallelization strategy

dlp_aocl_enable_instruction_query

Query AOCL_DLP_ENABLE_INSTRUCTIONS environment setting

dlp_version_query

Query library version (major, minor, patch)

API Selection Guide#

Choose the Right GEMM Variant#

By Precision Requirements:

  1. High Precision: f32f32f32of32 for maximum accuracy

  2. Balanced: bf16bf16f32of32 for good accuracy with reduced memory

  3. Quantized: u8s8s32os32 or s8s8s32os8 for inference

By Performance Needs:

  1. Single Operation: Standard GEMM functions

  2. Multiple Operations: Batch GEMM functions

  3. Repeated Operations: Use matrix reordering

Data Type Naming Convention#

Function names follow the pattern: [input_A][input_B][accumulation]o[output]

  • f32 = float32

  • bf16 = bfloat16

  • u8 = uint8

  • s8 = int8

  • s32 = int32

Example: bf16bf16f32of32 = bfloat16 inputs, float32 accumulation and output

See Also#

Indices and Tables#