diff options
Diffstat (limited to 'VA/dpu/task.c')
| -rw-r--r-- | VA/dpu/task.c | 109 |
1 files changed, 62 insertions, 47 deletions
diff --git a/VA/dpu/task.c b/VA/dpu/task.c index bb41303..9622911 100644 --- a/VA/dpu/task.c +++ b/VA/dpu/task.c @@ -15,10 +15,11 @@ __host dpu_arguments_t DPU_INPUT_ARGUMENTS; // vector_addition: Computes the vector addition of a cached block -static void vector_addition(T *bufferB, T *bufferA, unsigned int l_size) { - for (unsigned int i = 0; i < l_size; i++){ - bufferB[i] += bufferA[i]; - } +static void vector_addition(T *bufferB, T *bufferA, unsigned int l_size) +{ + for (unsigned int i = 0; i < l_size; i++) { + bufferB[i] += bufferA[i]; + } } // Barrier @@ -26,53 +27,67 @@ BARRIER_INIT(my_barrier, NR_TASKLETS); extern int main_kernel1(void); -int (*kernels[nr_kernels])(void) = {main_kernel1}; +int (*kernels[nr_kernels])(void) = { main_kernel1 }; -int main(void) { - // Kernel - return kernels[DPU_INPUT_ARGUMENTS.kernel](); +int main(void) +{ + // Kernel + return kernels[DPU_INPUT_ARGUMENTS.kernel] (); } // main_kernel1 -int main_kernel1() { - unsigned int tasklet_id = me(); +int main_kernel1() +{ + unsigned int tasklet_id = me(); #if PRINT - printf("tasklet_id = %u\n", tasklet_id); + printf("tasklet_id = %u\n", tasklet_id); #endif - if (tasklet_id == 0){ // Initialize once the cycle counter - mem_reset(); // Reset the heap - } - // Barrier - barrier_wait(&my_barrier); - - uint32_t input_size_dpu_bytes = DPU_INPUT_ARGUMENTS.size; // Input size per DPU in bytes - uint32_t input_size_dpu_bytes_transfer = DPU_INPUT_ARGUMENTS.transfer_size; // Transfer input size per DPU in bytes - - // Address of the current processing block in MRAM - uint32_t base_tasklet = tasklet_id << BLOCK_SIZE_LOG2; - uint32_t mram_base_addr_A = (uint32_t)DPU_MRAM_HEAP_POINTER; - uint32_t mram_base_addr_B = (uint32_t)(DPU_MRAM_HEAP_POINTER + input_size_dpu_bytes_transfer); - - // Initialize a local cache to store the MRAM block - T *cache_A = (T *) mem_alloc(BLOCK_SIZE); - T *cache_B = (T *) mem_alloc(BLOCK_SIZE); - - for(unsigned int byte_index = base_tasklet; byte_index < input_size_dpu_bytes; byte_index += BLOCK_SIZE * NR_TASKLETS){ - - // Bound checking - uint32_t l_size_bytes = (byte_index + BLOCK_SIZE >= input_size_dpu_bytes) ? (input_size_dpu_bytes - byte_index) : BLOCK_SIZE; - - // Load cache with current MRAM block - mram_read((__mram_ptr void const*)(mram_base_addr_A + byte_index), cache_A, l_size_bytes); - mram_read((__mram_ptr void const*)(mram_base_addr_B + byte_index), cache_B, l_size_bytes); - - // Computer vector addition - vector_addition(cache_B, cache_A, l_size_bytes >> DIV); - - // Write cache to current MRAM block - mram_write(cache_B, (__mram_ptr void*)(mram_base_addr_B + byte_index), l_size_bytes); - - } - - return 0; + if (tasklet_id == 0) { // Initialize once the cycle counter + mem_reset(); // Reset the heap + } + // Barrier + barrier_wait(&my_barrier); + + uint32_t input_size_dpu_bytes = DPU_INPUT_ARGUMENTS.size; // Input size per DPU in bytes + uint32_t input_size_dpu_bytes_transfer = DPU_INPUT_ARGUMENTS.transfer_size; // Transfer input size per DPU in bytes + + // Address of the current processing block in MRAM + uint32_t base_tasklet = tasklet_id << BLOCK_SIZE_LOG2; + uint32_t mram_base_addr_A = (uint32_t) DPU_MRAM_HEAP_POINTER; + uint32_t mram_base_addr_B = + (uint32_t) (DPU_MRAM_HEAP_POINTER + input_size_dpu_bytes_transfer); + + // Initialize a local cache to store the MRAM block + T *cache_A = (T *) mem_alloc(BLOCK_SIZE); + T *cache_B = (T *) mem_alloc(BLOCK_SIZE); + + for (unsigned int byte_index = base_tasklet; + byte_index < input_size_dpu_bytes; + byte_index += BLOCK_SIZE * NR_TASKLETS) { + + // Bound checking + uint32_t l_size_bytes = + (byte_index + BLOCK_SIZE >= + input_size_dpu_bytes) ? (input_size_dpu_bytes - + byte_index) : BLOCK_SIZE; + + // Load cache with current MRAM block + mram_read((__mram_ptr void const *)(mram_base_addr_A + + byte_index), cache_A, + l_size_bytes); + mram_read((__mram_ptr void const *)(mram_base_addr_B + + byte_index), cache_B, + l_size_bytes); + + // Computer vector addition + vector_addition(cache_B, cache_A, l_size_bytes >> DIV); + + // Write cache to current MRAM block + mram_write(cache_B, + (__mram_ptr void *)(mram_base_addr_B + byte_index), + l_size_bytes); + + } + + return 0; } |