Programming Examples

• DRAM Loopback
  • Silicon accelerator setup
  • Program pre-compilation setup
  • Building a data movement kernel
  • Create buffers in DRAM and L1
  • Sending real data into DRAM
  • Setting runtime arguments for the data movement kernel
  • Running the program
  • Launch and verify output
  • Validation and teardown
• Eltwise SFPU
  • Circular buffers for data movement to/from compute engine
  • Compile-time compute kernel arguments
  • Compute kernel declaration and compile-time defines
  • Extra runtime arguments for reader/writer
  • Conclusion
• Eltwise binary
  • New buffers
  • Compute kernel declaration and compile-time defines
  • Extra source tensor
  • Conclusion
• Matmul (Single Core)
  • Host Code
  • Main blocks in matmul_single_core function
  • Create Program, Enqueue initialization, and core range definition
  • Create DRAM buffers & Circular buffers
  • Compile-time kernels arguments
  • Compute kernel declaration and compile-time defines
  • Runtime arguments and program launch
  • Conclusion
• Matmul (Multi Core)
  • Accessing all the cores
  • Splitting the work across cores
  • Using different kernels for reader/writer
  • Compute kernel args
  • Reader/writer kernel runtime args
  • Conclusion
• Matmul (Multi Core Optimized)
  • Data Reuse in matmul_multicore_reuse
    • Fine-Grained Block Size Control
    • Intermediate Circular Buffer Configuration
    • Stride Kernel Arguments
    • Intermediate Results Handling
    • Conclusion
  • Data Multicasting in matmul_multicore_reuse_mcast
    • Additional Compile-Time Argument
    • Configuring Core Ranges for Tile Distribution
    • Circular Buffer Creation for CoreGrid
    • Multicast Reader/Writer Kernel Setup
    • New Compute Kernel: Fused Bias Addition and Activation Functions
    • Semaphores
    • Kernel Runtime Arguments