blob: 82cb18ede98f20746dd5b86bfbd29784315b43f8 [file] [log] [blame]
/*
* Copyright (c) 2021 Arm Limited. All rights reserved.
* SPDX-License-Identifier: Apache-2.0
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
__STACK_SIZE = 0x00060000;
__HEAP_SIZE = 0x000C0000;
/* System memory brief */
MEMORY
{
ITCM (rx) : ORIGIN = 0x00000000, LENGTH = 0x00080000
DTCM (rwx) : ORIGIN = 0x20000000, LENGTH = 0x00080000
BRAM (rwx) : ORIGIN = 0x11000000, LENGTH = 0x00100000
SRAM (rwx) : ORIGIN = 0x31000000, LENGTH = 0x00200000
DDR (rwx) : ORIGIN = 0x70000000, LENGTH = 0x02000000
}
/* Linker script to place sections and symbol values. Should be used together
* with other linker script that defines memory regions ITCM and RAM.
* It references following symbols, which must be defined in code:
* Reset_Handler : Entry of reset handler
*
* It defines following symbols, which code can use without definition:
* __exidx_start
* __exidx_end
* __copy_table_start__
* __copy_table_end__
* __zero_table_start__
* __zero_table_end__
* __etext
* __data_start__
* __preinit_array_start
* __preinit_array_end
* __init_array_start
* __init_array_end
* __fini_array_start
* __fini_array_end
* __data_end__
* __bss_start__
* __bss_end__
* __end__
* end
* __HeapLimit
* __StackLimit
* __StackTop
* __stack
*/
ENTRY(Reset_Handler)
SECTIONS
{
.text.at_itcm :
{
KEEP(*(.vectors))
/**
* All code goes here, with one exception of
* all_ops_resolver object file. This code
* instead placed on BRAM. See comment in the
* BRAM section for details.
**/
*(EXCLUDE_FILE(*all_ops_resolver.o *hal.c.obj) .text*)
KEEP(*(.init))
KEEP(*(.fini))
/* .ctors */
*crtbegin.o(.ctors)
*crtbegin?.o(.ctors)
*(EXCLUDE_FILE(*crtend?.o *crtend.o) .ctors)
*(SORT(.ctors.*))
*(.ctors)
/* .dtors */
*crtbegin.o(.dtors)
*crtbegin?.o(.dtors)
*(EXCLUDE_FILE(*crtend?.o *crtend.o) .dtors)
*(SORT(.dtors.*))
*(.dtors)
KEEP(*(.eh_frame*))
} > ITCM
__exidx_start = .;
.ARM.exidx.at_itcm :
{
*(.ARM.exidx* .gnu.linkonce.armexidx.*)
} > ITCM
__exidx_end = .;
.zero.table.at_itcm :
{
. = ALIGN(4);
__zero_table_start__ = .;
LONG (__bss_start__)
LONG ((__bss_end__ - __bss_start__)/4) /* Size is in 32-bit words */
__zero_table_end__ = .;
} > ITCM
.copy.table.at_itcm :
{
. = ALIGN(4);
__copy_table_start__ = .;
/* Section to be copied - part 1: any data to be placed in BRAM */
LONG (__etext)
LONG (__data_start__)
LONG ((__data_end__ - __data_start__)/4) /* Size is in 32-bit words */
/* Section to be copied - part 2: RO data for for DTCM */
LONG (__etext2)
LONG (__ro_data_start__)
LONG ((__ro_data_end__ - __ro_data_start__)/4) /* Size is in 32-bit words */
__copy_table_end__ = .;
} > ITCM
__itcm_total = ALIGN(4);
ASSERT( __itcm_total < (ORIGIN(ITCM) + LENGTH(ITCM)), "ITCM overflow")
.sram :
{
. = ALIGN(16);
/* Cache area (if used) */
*(.bss.NoInit.ethos_u_cache)
. = ALIGN (16);
/* activation buffers a.k.a tensor arena when memory mode sram only or shared sram */
*(.bss.NoInit.activation_buf_sram)
. = ALIGN(16);
} > SRAM AT > SRAM
.bss :
{
. = ALIGN(4);
__bss_start__ = .;
*(.bss)
*(.bss.*)
*(COMMON)
. = ALIGN(4);
__bss_end__ = .;
} > DTCM AT > DTCM
.stack (ORIGIN(DTCM) + LENGTH(DTCM) - __STACK_SIZE) (COPY) :
{
. = ALIGN(8);
__StackLimit = .;
. = . + __STACK_SIZE;
. = ALIGN(8);
__StackTop = .;
} > DTCM
PROVIDE(__stack = __StackTop);
ASSERT(
(__STACK_SIZE + __bss_end__ - __bss_start__) <= LENGTH(DTCM),
"DTCM overflow")
.ddr.at_ddr :
{
/* __attribute__((aligned(16))) is not handled by the CMSIS startup code.
* Force the alignment here as a workaround */
. = ALIGN(16);
/* nn model's baked in input matrices */
*(ifm)
. = ALIGN(16);
/* nn model's default space */
*(nn_model)
. = ALIGN (16);
/* labels */
*(labels)
. = ALIGN (16);
/* activation buffers a.k.a tensor arena when memory mode dedicated sram */
*(activation_buf_dram)
. = ALIGN (16);
} > DDR AT > DDR
/**
* Location counter can end up 2byte aligned with narrow Thumb code but
* __etext is assumed by startup code to be the LMA of a section in DTCM
* which must be 4byte aligned
*/
__etext = ALIGN (4);
.bram.at_ddr : AT (__etext)
{
__data_start__ = .;
*(vtable)
*(.data)
*(.data.*)
. = ALIGN(4);
PROVIDE_HIDDEN (__preinit_array_start = .);
KEEP(*(.preinit_array))
PROVIDE_HIDDEN (__preinit_array_end = .);
. = ALIGN(4);
PROVIDE_HIDDEN (__init_array_start = .);
KEEP(*(SORT(.init_array.*)))
KEEP(*(.init_array))
PROVIDE_HIDDEN (__init_array_end = .);
. = ALIGN(4);
PROVIDE_HIDDEN (__fini_array_start = .);
KEEP(*(SORT(.fini_array.*)))
KEEP(*(.fini_array))
PROVIDE_HIDDEN (__fini_array_end = .);
KEEP(*(.jcr*))
. = ALIGN(4);
*(.ARM.extab* .gnu.linkonce.armextab.*)
. = ALIGN(4);
/**
* Place the all ops resolver code data here. This accounts
* for ~4k worth of saving on the ITCM load region. It is
* only designed to be included (by default) for the inference
* runner use case.
**/
*all_ops_resolver.o (*.text*)
. = ALIGN(4);
*hal.c.obj (*.text*)
. = ALIGN(4);
__data_end__ = .;
} > BRAM
__etext2 = __etext + (__data_end__ - __data_start__);
.data.at_ddr : AT (__etext2)
{
. = ALIGN(4);
__ro_data_start__ = .;
*(.rodata*)
. = ALIGN(4);
* (npu_driver_version)
. = ALIGN(4);
* (npu_driver_arch_version)
. = ALIGN(4);
__ro_data_end__ = .;
} > BRAM
.heap (COPY) :
{
. = ALIGN(8);
__end__ = .;
PROVIDE(end = .);
. = . + __HEAP_SIZE;
. = ALIGN(8);
__HeapLimit = .;
} > BRAM
ASSERT (
(__ro_data_end__ - __ro_data_start__)
+ (__data_end__ - __data_start__)
+ __HEAP_SIZE <= LENGTH(BRAM),
"BRAM overflow")
}