diff options
Diffstat (limited to 'ethosu/vela/compiler_driver.py')
| -rw-r--r-- | ethosu/vela/compiler_driver.py | 35 |
1 files changed, 17 insertions, 18 deletions
diff --git a/ethosu/vela/compiler_driver.py b/ethosu/vela/compiler_driver.py index 9e1cb3ab..0739133b 100644 --- a/ethosu/vela/compiler_driver.py +++ b/ethosu/vela/compiler_driver.py @@ -225,31 +225,18 @@ def compiler_driver(nng, arch, options, scheduler_options): root_sg = nng.get_root_subgraph() alloc_list = [] - feature_maps_in_fast_storage = arch.feature_map_storage_mem_area == arch.fast_storage_mem_area - if feature_maps_in_fast_storage: - mem_alloc_scratch = (arch.feature_map_storage_mem_area, set((MemType.Scratch, MemType.Scratch_fast))) - alloc_list.append(mem_alloc_scratch) - else: + if arch.is_spilling_enabled(): mem_alloc_scratch_fast = (arch.fast_storage_mem_area, set((MemType.Scratch_fast,))) mem_alloc_scratch = (arch.feature_map_storage_mem_area, set((MemType.Scratch,))) # Order is important alloc_list.append(mem_alloc_scratch_fast) alloc_list.append(mem_alloc_scratch) + else: + mem_alloc_scratch = (arch.feature_map_storage_mem_area, set((MemType.Scratch, MemType.Scratch_fast))) + alloc_list.append(mem_alloc_scratch) for mem_area, mem_type_set in alloc_list: - if feature_maps_in_fast_storage or mem_area != arch.fast_storage_mem_area: - tensor_allocation.allocate_tensors( - nng, - root_sg, - arch, - mem_area, - mem_type_set, - tensor_allocator=options.tensor_allocator, - verbose_allocation=options.verbose_allocation, - show_minimum_possible_allocation=options.show_minimum_possible_allocation, - allocation_alignment=options.allocation_alignment, - ) - else: + if arch.is_spilling_enabled() and mem_area == arch.fast_storage_mem_area: # For the case where scratch_fast != scratch: attempt to place feature maps used between # cascaded passes in fast storage. Bisection is used to find the max possible usage of SRAM. alloc_results = [] @@ -285,6 +272,18 @@ def compiler_driver(nng, arch, options, scheduler_options): "Increasing the value of --weight-estimation-scaling may help to resolve the issue. " "See OPTIONS.md for more information.".format(arch.sram_size) ) + else: + tensor_allocation.allocate_tensors( + nng, + root_sg, + arch, + mem_area, + mem_type_set, + tensor_allocator=options.tensor_allocator, + verbose_allocation=options.verbose_allocation, + show_minimum_possible_allocation=options.show_minimum_possible_allocation, + allocation_alignment=options.allocation_alignment, + ) # Generate command streams and serialise Npu-ops into tensors for sg in nng.subgraphs: |