Module shared_memory 

Shared memory management for CUDA kernel emulation

Emulates CUDA shared memory (__shared__) on the CPU, including:

• Static shared memory allocation (known size at compile time)
• Dynamic (extern) shared memory allocation (size provided at launch)
• Bank conflict detection for profiling/debugging

In CUDA, shared memory is per-block SRAM accessible by all threads in a block. On the CPU we emulate this with heap-allocated buffers shared among threads in the same block.

Structs

BankConflictDetector

Tracks shared memory access patterns to detect bank conflicts.

DynamicSharedMemory

Dynamic shared memory allocation.

SharedMemory

Static shared memory allocation.

Constants

BANK_WIDTH_BYTES

Size of each bank in bytes (4 bytes = 32 bits, matching CUDA).

NUM_BANKS

Number of memory banks (matches NVIDIA GPU shared memory banks).