riscv
/
OpenFPGA
mirror of https://github.com/lnis-uofu/OpenFPGA.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #6 from RapidSilicon/phy_mem_bank 

Alpha Version of New Configuration Protocol: Physical Design Friendly Memory Bank
This commit is contained in:
tangxifan 2021-09-10 21:18:32 -07:00 committed by GitHub
parent c206c4e95e 4af6413c97
commit d0e60c0697
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
38 changed files with 1790 additions and 102 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
docs/source/manual/arch_lang/config_protocol.rst
View File

@ -16,7 +16,7 @@ Template
<organization type="<string>" circuit_model_name="<string>" num_regions="<int>"/> <organization type="<string>" circuit_model_name="<string>" num_regions="<int>"/>
</configuration_protocol> </configuration_protocol>
.. option:: type="scan_chain|memory_bank|standalone" .. option:: type="scan_chain|memory_bank|standalone|frame_based|ql_memory_bank"
Specify the type of configuration circuits. Specify the type of configuration circuits.
@ -24,6 +24,7 @@ Template
- ``scan_chain``: configurable memories are connected in a chain. Bitstream is loaded serially to program a FPGA - ``scan_chain``: configurable memories are connected in a chain. Bitstream is loaded serially to program a FPGA
- ``frame_based``: configurable memories are organized by frames. Each module of a FPGA fabric, e.g., Configurable Logic Block (CLB), Switch Block (SB) and Connection Block (CB), is considered as a frame of configurable memories. Inside each frame, all the memory banks are accessed through an address decoder. Users can write each memory cell with a specific address. Note that the frame-based memory organization is applid hierarchically. Each frame may consists of a number of sub frames, each of which follows the similar organization. - ``frame_based``: configurable memories are organized by frames. Each module of a FPGA fabric, e.g., Configurable Logic Block (CLB), Switch Block (SB) and Connection Block (CB), is considered as a frame of configurable memories. Inside each frame, all the memory banks are accessed through an address decoder. Users can write each memory cell with a specific address. Note that the frame-based memory organization is applid hierarchically. Each frame may consists of a number of sub frames, each of which follows the similar organization.
- ``memory_bank``: configurable memories are organized in an array, where each element can be accessed by an unique address to the BL/WL decoders - ``memory_bank``: configurable memories are organized in an array, where each element can be accessed by an unique address to the BL/WL decoders
- ``ql_memory_bank``: configurable memories are organized in an array, where each element can be accessed by an unique address to the BL/WL decoders. This is a physical design friendly memory bank organization, where BL/WLs are efficiently shared by programmable blocks per column and row
- ``standalone``: configurable memories are directly accessed through ports of FPGA fabrics. In other words, there are no protocol to control the memories. This allows full customization on the configuration protocol for hardware engineers. - ``standalone``: configurable memories are directly accessed through ports of FPGA fabrics. In other words, there are no protocol to control the memories. This allows full customization on the configuration protocol for hardware engineers.
.. note:: Avoid to use ``standalone`` when designing an FPGA chip. It will causes a huge number of I/Os required, far beyond any package size. It is well applicable to eFPGAs, where designers do need customized protocols between FPGA and processors. .. note:: Avoid to use ``standalone`` when designing an FPGA chip. It will causes a huge number of I/Os required, far beyond any package size. It is well applicable to eFPGAs, where designers do need customized protocols between FPGA and processors.
@ -147,6 +148,26 @@ Users can customized the number of memory banks to be used across the fabrics. B
.. warning:: Please do NOT add inverted Bit-Line and Word-Line inputs. It is not supported yet! .. warning:: Please do NOT add inverted Bit-Line and Word-Line inputs. It is not supported yet!
QuickLogic Memory bank Example
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The following XML code describes a physical design friendly memory-bank circuitry to configure the core logic of FPGA, as illustrated in :numref:`fig_memory_bank`.
It will use the circuit model defined in :numref:`fig_sram_blwl`.
.. code-block:: xml
<configuration_protocol>
<organization type="ql_memory_bank" circuit_model_name="sram_blwl"/>
</configuration_protocol>
.. note:: Memory-bank decoders does require a memory cell to have
- two outputs (one regular and another inverted)
- a Bit-Line input to load the data
- a Word-Line input to enable data write
.. warning:: Please do NOT add inverted Bit-Line and Word-Line inputs. It is not supported yet!
Standalone SRAM Example Standalone SRAM Example
~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~
In the standalone configuration protocol, every memory cell of the core logic of a FPGA fabric can be directly accessed at the top-level module, as illustrated in :numref:`fig_vanilla_config_protocol`. In the standalone configuration protocol, every memory cell of the core logic of a FPGA fabric can be directly accessed at the top-level module, as illustrated in :numref:`fig_vanilla_config_protocol`.

3
libopenfpga/libarchopenfpga/src/circuit_types.h
View File

@ -125,10 +125,11 @@ enum e_config_protocol_type {
CONFIG_MEM_STANDALONE, CONFIG_MEM_STANDALONE,
CONFIG_MEM_SCAN_CHAIN, CONFIG_MEM_SCAN_CHAIN,
CONFIG_MEM_MEMORY_BANK, CONFIG_MEM_MEMORY_BANK,
CONFIG_MEM_QL_MEMORY_BANK,
CONFIG_MEM_FRAME_BASED, CONFIG_MEM_FRAME_BASED,
NUM_CONFIG_PROTOCOL_TYPES NUM_CONFIG_PROTOCOL_TYPES
}; };
constexpr std::array<const char*, NUM_CONFIG_PROTOCOL_TYPES> CONFIG_PROTOCOL_TYPE_STRING = {{"standalone", "scan_chain", "memory_bank", "frame_based"}}; constexpr std::array<const char*, NUM_CONFIG_PROTOCOL_TYPES> CONFIG_PROTOCOL_TYPE_STRING = {{"standalone", "scan_chain", "memory_bank", "ql_memory_bank", "frame_based"}};
#endif #endif

1
openfpga/src/base/openfpga_bitstream.cpp
View File

@ -76,6 +76,7 @@ int build_fabric_bitstream(OpenfpgaContext& openfpga_ctx,
/* Build fabric bitstream here */ /* Build fabric bitstream here */
openfpga_ctx.mutable_fabric_bitstream() = build_fabric_dependent_bitstream(openfpga_ctx.bitstream_manager(), openfpga_ctx.mutable_fabric_bitstream() = build_fabric_dependent_bitstream(openfpga_ctx.bitstream_manager(),
openfpga_ctx.module_graph(), openfpga_ctx.module_graph(),
openfpga_ctx.arch().circuit_lib,
openfpga_ctx.arch().config_protocol, openfpga_ctx.arch().config_protocol,
cmd_context.option_enable(cmd, opt_verbose)); cmd_context.option_enable(cmd, opt_verbose));

1
openfpga/src/base/openfpga_naming.cpp
View File

@ -703,6 +703,7 @@ std::string generate_sram_port_name(const e_config_protocol_type& sram_orgz_type
} }
break; break;
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
/* Two types of ports are available: /* Two types of ports are available:
* (1) Bit Lines (BLs) of a SRAM cell, enabled by port type of BL * (1) Bit Lines (BLs) of a SRAM cell, enabled by port type of BL

2
openfpga/src/fabric/build_grid_modules.cpp
View File

@ -279,7 +279,7 @@ void build_primitive_block_module(ModuleManager& module_manager,
size_t num_shared_config_bits = find_circuit_num_shared_config_bits(circuit_lib, primitive_model, sram_orgz_type); size_t num_shared_config_bits = find_circuit_num_shared_config_bits(circuit_lib, primitive_model, sram_orgz_type);
if (0 < num_shared_config_bits) { if (0 < num_shared_config_bits) {
/* Check: this SRAM organization type must be memory-bank ! */ /* Check: this SRAM organization type must be memory-bank ! */
VTR_ASSERT( CONFIG_MEM_MEMORY_BANK == sram_orgz_type ); VTR_ASSERT( CONFIG_MEM_MEMORY_BANK == sram_orgz_type || CONFIG_MEM_QL_MEMORY_BANK == sram_orgz_type );
/* Generate a list of ports */ /* Generate a list of ports */
add_reserved_sram_ports_to_module_manager(module_manager, primitive_module, add_reserved_sram_ports_to_module_manager(module_manager, primitive_module,
num_shared_config_bits); num_shared_config_bits);

1
openfpga/src/fabric/build_memory_modules.cpp
View File

@ -786,6 +786,7 @@ void build_memory_module(ModuleManager& module_manager,
const size_t& num_mems) { const size_t& num_mems) {
switch (sram_orgz_type) { switch (sram_orgz_type) {
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
build_memory_flatten_module(module_manager, circuit_lib, build_memory_flatten_module(module_manager, circuit_lib,
module_name, sram_model, num_mems); module_name, sram_model, num_mems);

3
openfpga/src/fabric/build_top_module.cpp
View File

@ -407,7 +407,7 @@ int build_top_module(ModuleManager& module_manager,
* This is a much easier job after adding sub modules (instances), * This is a much easier job after adding sub modules (instances),
* we just need to find all the I/O ports from the child modules and build a list of it * we just need to find all the I/O ports from the child modules and build a list of it
*/ */
vtr::vector<ConfigRegionId, size_t> top_module_num_config_bits = find_top_module_regional_num_config_bit(module_manager, top_module, circuit_lib, sram_model, config_protocol.type()); TopModuleNumConfigBits top_module_num_config_bits = find_top_module_regional_num_config_bit(module_manager, top_module, circuit_lib, sram_model, config_protocol.type());
if (!top_module_num_config_bits.empty()) { if (!top_module_num_config_bits.empty()) {
add_top_module_sram_ports(module_manager, top_module, add_top_module_sram_ports(module_manager, top_module,
@ -422,6 +422,7 @@ int build_top_module(ModuleManager& module_manager,
if (0 < module_manager.configurable_children(top_module).size()) { if (0 < module_manager.configurable_children(top_module).size()) {
add_top_module_nets_memory_config_bus(module_manager, decoder_lib, add_top_module_nets_memory_config_bus(module_manager, decoder_lib,
top_module, top_module,
circuit_lib, sram_model,
config_protocol, circuit_lib.design_tech_type(sram_model), config_protocol, circuit_lib.design_tech_type(sram_model),
top_module_num_config_bits); top_module_num_config_bits);
} }

211
openfpga/src/fabric/build_top_module_memory.cpp
View File

@ -20,9 +20,11 @@
#include "openfpga_naming.h" #include "openfpga_naming.h"
#include "memory_utils.h" #include "memory_utils.h"
#include "memory_bank_utils.h"
#include "decoder_library_utils.h" #include "decoder_library_utils.h"
#include "module_manager_utils.h" #include "module_manager_utils.h"
#include "build_decoder_modules.h" #include "build_decoder_modules.h"
#include "build_top_module_memory_bank.h"
#include "build_top_module_memory.h" #include "build_top_module_memory.h"
/* begin namespace openfpga */ /* begin namespace openfpga */
@ -74,8 +76,14 @@ void organize_top_module_tile_cb_modules(ModuleManager& module_manager,
if (0 < find_module_num_config_bits(module_manager, cb_module, if (0 < find_module_num_config_bits(module_manager, cb_module,
circuit_lib, sram_model, circuit_lib, sram_model,
sram_orgz_type)) { sram_orgz_type)) {
/* CBX coordinate conversion calculation: (1,0) -> (2,1) */
vtr::Point<int> config_coord(rr_gsb.get_cb_x(cb_type) * 2, rr_gsb.get_cb_y(cb_type) * 2 + 1);
if (cb_type == CHANY) {
/* CBY has a different coordinate conversion calculation: (0,1) -> (1,2) */
config_coord.set(rr_gsb.get_cb_x(cb_type) * 2 + 1, rr_gsb.get_cb_y(cb_type) * 2);
}
/* Note that use the original CB coodinate for instance id searching ! */ /* Note that use the original CB coodinate for instance id searching ! */
module_manager.add_configurable_child(top_module, cb_module, cb_instance_ids[rr_gsb.get_cb_x(cb_type)][rr_gsb.get_cb_y(cb_type)]); module_manager.add_configurable_child(top_module, cb_module, cb_instance_ids[rr_gsb.get_cb_x(cb_type)][rr_gsb.get_cb_y(cb_type)], config_coord);
} }
} }
@ -84,6 +92,37 @@ void organize_top_module_tile_cb_modules(ModuleManager& module_manager,
* to the memory modules and memory instances * to the memory modules and memory instances
* This function is designed for organizing memory modules in top-level * This function is designed for organizing memory modules in top-level
* module * module
* This function also adds coordindates for each configurable child under the top-level module
* of a FPGA fabric. A configurable child could be a programmable block (grid),
* a Connection Block (CBx/y) or a Switch block (SB).
* This function, we consider a coordinate system as follows
* - Each row may consist of either (1) grid and CBy or (2) CBx and SB
* - Each column may consist of either (1) grid and CBx or (2) CBy and SB
*
* Column 0 Column 1
*
* +---------------+----------+
* | | |
* | | |
* | Grid | CBY | Row 3
* | | |
* | | |
* +---------------+----------+
* | | |
* | CBX | SB | Row 2
* | | |
* +---------------+----------+
* | | |
* | | |
* | Grid | CBY | Row 1
* | | |
* | | |
* +---------------+----------+
* | | |
* | CBX | SB | Row 0
* | | |
* +---------------+----------+
*******************************************************************/ *******************************************************************/
static static
void organize_top_module_tile_memory_modules(ModuleManager& module_manager, void organize_top_module_tile_memory_modules(ModuleManager& module_manager,
@ -130,7 +169,8 @@ void organize_top_module_tile_memory_modules(ModuleManager& module_manager,
if (0 < find_module_num_config_bits(module_manager, sb_module, if (0 < find_module_num_config_bits(module_manager, sb_module,
circuit_lib, sram_model, circuit_lib, sram_model,
sram_orgz_type)) { sram_orgz_type)) {
module_manager.add_configurable_child(top_module, sb_module, sb_instance_ids[rr_gsb.get_sb_x()][rr_gsb.get_sb_y()]); vtr::Point<int> config_coord(rr_gsb.get_sb_x() * 2 + 1, rr_gsb.get_sb_y() * 2 + 1);
module_manager.add_configurable_child(top_module, sb_module, sb_instance_ids[rr_gsb.get_sb_x()][rr_gsb.get_sb_y()], config_coord);
} }
} }
@ -172,11 +212,11 @@ void organize_top_module_tile_memory_modules(ModuleManager& module_manager,
if (0 < find_module_num_config_bits(module_manager, grid_module, if (0 < find_module_num_config_bits(module_manager, grid_module,
circuit_lib, sram_model, circuit_lib, sram_model,
sram_orgz_type)) { sram_orgz_type)) {
module_manager.add_configurable_child(top_module, grid_module, grid_instance_ids[tile_coord.x()][tile_coord.y()]); vtr::Point<int> config_coord(tile_coord.x() * 2, tile_coord.y() * 2);
module_manager.add_configurable_child(top_module, grid_module, grid_instance_ids[tile_coord.x()][tile_coord.y()], config_coord);
} }
} }
/******************************************************************** /********************************************************************
* Split memory modules into different configurable regions * Split memory modules into different configurable regions
* This function will create regions based on the definition * This function will create regions based on the definition
@ -231,7 +271,8 @@ void build_top_module_configurable_regions(ModuleManager& module_manager,
/* Exclude decoders from the list */ /* Exclude decoders from the list */
size_t num_configurable_children = module_manager.configurable_children(top_module).size(); size_t num_configurable_children = module_manager.configurable_children(top_module).size();
if (CONFIG_MEM_MEMORY_BANK == config_protocol.type()) { if (CONFIG_MEM_MEMORY_BANK == config_protocol.type()
|| CONFIG_MEM_QL_MEMORY_BANK == config_protocol.type()) {
num_configurable_children -= 2; num_configurable_children -= 2;
} else if (CONFIG_MEM_FRAME_BASED == config_protocol.type()) { } else if (CONFIG_MEM_FRAME_BASED == config_protocol.type()) {
num_configurable_children -= 1; num_configurable_children -= 1;
@ -496,6 +537,7 @@ void shuffle_top_module_configurable_children(ModuleManager& module_manager,
/* Cache the configurable children and their instances */ /* Cache the configurable children and their instances */
std::vector<ModuleId> orig_configurable_children = module_manager.configurable_children(top_module); std::vector<ModuleId> orig_configurable_children = module_manager.configurable_children(top_module);
std::vector<size_t> orig_configurable_child_instances = module_manager.configurable_child_instances(top_module); std::vector<size_t> orig_configurable_child_instances = module_manager.configurable_child_instances(top_module);
std::vector<vtr::Point<int>> orig_configurable_child_coordinates = module_manager.configurable_child_coordinates(top_module);
/* Reorganize the configurable children */ /* Reorganize the configurable children */
module_manager.clear_configurable_children(top_module); module_manager.clear_configurable_children(top_module);
@ -503,7 +545,8 @@ void shuffle_top_module_configurable_children(ModuleManager& module_manager,
for (size_t ikey = 0; ikey < num_keys; ++ikey) { for (size_t ikey = 0; ikey < num_keys; ++ikey) {
module_manager.add_configurable_child(top_module, module_manager.add_configurable_child(top_module,
orig_configurable_children[shuffled_keys[ikey]], orig_configurable_children[shuffled_keys[ikey]],
orig_configurable_child_instances[shuffled_keys[ikey]]); orig_configurable_child_instances[shuffled_keys[ikey]],
orig_configurable_child_coordinates[shuffled_keys[ikey]]);
} }
/* Reset configurable regions */ /* Reset configurable regions */
@ -614,13 +657,13 @@ int load_top_module_memory_modules_from_fabric_key(ModuleManager& module_manager
* - This function should be called after the configurable children * - This function should be called after the configurable children
* is loaded to the top-level module! * is loaded to the top-level module!
********************************************************************/ ********************************************************************/
vtr::vector<ConfigRegionId, size_t> find_top_module_regional_num_config_bit(const ModuleManager& module_manager, TopModuleNumConfigBits find_top_module_regional_num_config_bit(const ModuleManager& module_manager,
const ModuleId& top_module, const ModuleId& top_module,
const CircuitLibrary& circuit_lib, const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model, const CircuitModelId& sram_model,
const e_config_protocol_type& config_protocol_type) { const e_config_protocol_type& config_protocol_type) {
/* Initialize the number of configuration bits for each region */ /* Initialize the number of configuration bits for each region */
vtr::vector<ConfigRegionId, size_t> num_config_bits(module_manager.regions(top_module).size(), 0); TopModuleNumConfigBits num_config_bits(module_manager.regions(top_module).size(), std::pair<size_t, size_t>(0, 0));
switch (config_protocol_type) { switch (config_protocol_type) {
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
@ -632,7 +675,35 @@ vtr::vector<ConfigRegionId, size_t> find_top_module_regional_num_config_bit(cons
*/ */
for (const ConfigRegionId& config_region : module_manager.regions(top_module)) { for (const ConfigRegionId& config_region : module_manager.regions(top_module)) {
for (const ModuleId& child_module : module_manager.region_configurable_children(top_module, config_region)) { for (const ModuleId& child_module : module_manager.region_configurable_children(top_module, config_region)) {
num_config_bits[config_region] += find_module_num_config_bits(module_manager, child_module, circuit_lib, sram_model, config_protocol_type); num_config_bits[config_region].first += find_module_num_config_bits(module_manager, child_module, circuit_lib, sram_model, config_protocol_type);
}
}
break;
}
case CONFIG_MEM_QL_MEMORY_BANK: {
/* For QL memory bank: we will use the row and column information for each configuration child
* in order to identify the number of unique BLs and WLs
* In this configuration protocol,
* - all the configurable child in the same row will share the same WLs
* - the number of WLs per row is limited by the configurable child which requires most WLs
* - each row has independent WLs
* - all the configurable child in the same column will share the same BLs
* - the number of BLs per column is limited by the configurable child which requires most BLs
* - each column has independent BLs
*/
for (const ConfigRegionId& config_region : module_manager.regions(top_module)) {
std::map<int, size_t> num_bls_per_tile = compute_memory_bank_regional_bitline_numbers_per_tile(module_manager, top_module,
config_region,
circuit_lib, sram_model);
std::map<int, size_t> num_wls_per_tile = compute_memory_bank_regional_wordline_numbers_per_tile(module_manager, top_module,
config_region,
circuit_lib, sram_model,
num_bls_per_tile);
for (const auto& kv : num_bls_per_tile) {
num_config_bits[config_region].first += kv.second;
}
for (const auto& kv : num_wls_per_tile) {
num_config_bits[config_region].second += kv.second;
} }
} }
break; break;
@ -646,14 +717,14 @@ vtr::vector<ConfigRegionId, size_t> find_top_module_regional_num_config_bit(cons
for (const ConfigRegionId& config_region : module_manager.regions(top_module)) { for (const ConfigRegionId& config_region : module_manager.regions(top_module)) {
for (const ModuleId& child_module : module_manager.region_configurable_children(top_module, config_region)) { for (const ModuleId& child_module : module_manager.region_configurable_children(top_module, config_region)) {
size_t temp_num_config_bits = find_module_num_config_bits(module_manager, child_module, circuit_lib, sram_model, config_protocol_type); size_t temp_num_config_bits = find_module_num_config_bits(module_manager, child_module, circuit_lib, sram_model, config_protocol_type);
num_config_bits[config_region] = std::max((int)temp_num_config_bits, (int)num_config_bits[config_region]); num_config_bits[config_region].first = std::max(temp_num_config_bits, num_config_bits[config_region].first);
} }
/* If there are more than 2 configurable children, we need a decoder /* If there are more than 2 configurable children, we need a decoder
* Otherwise, we can just short wire the address port to the children * Otherwise, we can just short wire the address port to the children
*/ */
if (1 < module_manager.region_configurable_children(top_module, config_region).size()) { if (1 < module_manager.region_configurable_children(top_module, config_region).size()) {
num_config_bits[config_region] += find_mux_local_decoder_addr_size(module_manager.region_configurable_children(top_module, config_region).size()); num_config_bits[config_region].first += find_mux_local_decoder_addr_size(module_manager.region_configurable_children(top_module, config_region).size());
} }
} }
@ -687,6 +758,7 @@ size_t generate_top_module_sram_port_size(const ConfigProtocol& config_protocol,
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
break; break;
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
case CONFIG_MEM_FRAME_BASED: case CONFIG_MEM_FRAME_BASED:
/* CCFF head/tail, data input could be multi-bit ports */ /* CCFF head/tail, data input could be multi-bit ports */
@ -706,34 +778,40 @@ size_t generate_top_module_sram_port_size(const ConfigProtocol& config_protocol,
* top-level module * top-level module
* The type and names of added ports strongly depend on the * The type and names of added ports strongly depend on the
* organization of SRAMs. * organization of SRAMs.
* 1. Standalone SRAMs: * - Standalone SRAMs:
* two ports will be added, which are BL and WL * two ports will be added, which are BL and WL
* 2. Scan-chain Flip-flops: * - Scan-chain Flip-flops:
* two ports will be added, which are the head of scan-chain * two ports will be added, which are the head of scan-chain
* and the tail of scan-chain * and the tail of scan-chain
* IMPORTANT: the port size will be forced to 1 in this case * IMPORTANT: the port size will be forced to 1 in this case
* because the head and tail are both 1-bit ports!!! * because the head and tail are both 1-bit ports!!!
* 3. Memory decoders: * - Memory decoders:
* - An enable signal * - An enable signal
* - A BL address port * - A BL address port
* - A WL address port * - A WL address port
* - A data-in port for the BL decoder * - A data-in port for the BL decoder
* 4. Frame-based memory: * - QL memory decoder:
* - An Enable signal * - An enable signal
* - An address port, whose size depends on the number of config bits * - An BL address port
* and the maximum size of address ports of configurable children * - A WL address port
* - An data_in port (single-bit) * - A data-in port for the BL decoder
* @note In this memory decoders, the address size will be computed in a different way than the regular one
* - Frame-based memory:
* - An Enable signal
* - An address port, whose size depends on the number of config bits
* and the maximum size of address ports of configurable children
* - An data_in port (single-bit)
********************************************************************/ ********************************************************************/
void add_top_module_sram_ports(ModuleManager& module_manager, void add_top_module_sram_ports(ModuleManager& module_manager,
const ModuleId& module_id, const ModuleId& module_id,
const CircuitLibrary& circuit_lib, const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model, const CircuitModelId& sram_model,
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const vtr::vector<ConfigRegionId, size_t>& num_config_bits) { const TopModuleNumConfigBits& num_config_bits) {
std::vector<std::string> sram_port_names = generate_sram_port_names(circuit_lib, sram_model, config_protocol.type()); std::vector<std::string> sram_port_names = generate_sram_port_names(circuit_lib, sram_model, config_protocol.type());
size_t total_num_config_bits = 0; size_t total_num_config_bits = 0;
for (const size_t& curr_num_config_bits : num_config_bits) { for (const auto& curr_num_config_bits : num_config_bits) {
total_num_config_bits += curr_num_config_bits; total_num_config_bits += curr_num_config_bits.first;
} }
size_t sram_port_size = generate_top_module_sram_port_size(config_protocol, total_num_config_bits); size_t sram_port_size = generate_top_module_sram_port_size(config_protocol, total_num_config_bits);
@ -754,7 +832,7 @@ void add_top_module_sram_ports(ModuleManager& module_manager,
/* BL address size is the largest among all the regions */ /* BL address size is the largest among all the regions */
size_t bl_addr_size = 0; size_t bl_addr_size = 0;
for (const ConfigRegionId& config_region : module_manager.regions(module_id)) { for (const ConfigRegionId& config_region : module_manager.regions(module_id)) {
bl_addr_size = std::max(bl_addr_size, find_memory_decoder_addr_size(num_config_bits[config_region])); bl_addr_size = std::max(bl_addr_size, find_memory_decoder_addr_size(num_config_bits[config_region].first));
} }
BasicPort bl_addr_port(std::string(DECODER_BL_ADDRESS_PORT_NAME), bl_addr_size); BasicPort bl_addr_port(std::string(DECODER_BL_ADDRESS_PORT_NAME), bl_addr_size);
module_manager.add_port(module_id, bl_addr_port, ModuleManager::MODULE_INPUT_PORT); module_manager.add_port(module_id, bl_addr_port, ModuleManager::MODULE_INPUT_PORT);
@ -762,7 +840,33 @@ void add_top_module_sram_ports(ModuleManager& module_manager,
/* WL address size is the largest among all the regions */ /* WL address size is the largest among all the regions */
size_t wl_addr_size = 0; size_t wl_addr_size = 0;
for (const ConfigRegionId& config_region : module_manager.regions(module_id)) { for (const ConfigRegionId& config_region : module_manager.regions(module_id)) {
wl_addr_size = std::max(wl_addr_size, find_memory_decoder_addr_size(num_config_bits[config_region])); wl_addr_size = std::max(wl_addr_size, find_memory_decoder_addr_size(num_config_bits[config_region].first));
}
BasicPort wl_addr_port(std::string(DECODER_WL_ADDRESS_PORT_NAME), wl_addr_size);
module_manager.add_port(module_id, wl_addr_port, ModuleManager::MODULE_INPUT_PORT);
/* Data input should be dependent on the number of configuration regions*/
BasicPort din_port(std::string(DECODER_DATA_IN_PORT_NAME), config_protocol.num_regions());
module_manager.add_port(module_id, din_port, ModuleManager::MODULE_INPUT_PORT);
break;
}
case CONFIG_MEM_QL_MEMORY_BANK: {
BasicPort en_port(std::string(DECODER_ENABLE_PORT_NAME), 1);
module_manager.add_port(module_id, en_port, ModuleManager::MODULE_INPUT_PORT);
/* BL address size is the largest among all the regions */
size_t bl_addr_size = 0;
for (const ConfigRegionId& config_region : module_manager.regions(module_id)) {
bl_addr_size = std::max(bl_addr_size, find_mux_local_decoder_addr_size(num_config_bits[config_region].first));
}
BasicPort bl_addr_port(std::string(DECODER_BL_ADDRESS_PORT_NAME), bl_addr_size);
module_manager.add_port(module_id, bl_addr_port, ModuleManager::MODULE_INPUT_PORT);
/* WL address size is the largest among all the regions */
size_t wl_addr_size = 0;
for (const ConfigRegionId& config_region : module_manager.regions(module_id)) {
wl_addr_size = std::max(wl_addr_size, find_mux_local_decoder_addr_size(num_config_bits[config_region].second));
} }
BasicPort wl_addr_port(std::string(DECODER_WL_ADDRESS_PORT_NAME), wl_addr_size); BasicPort wl_addr_port(std::string(DECODER_WL_ADDRESS_PORT_NAME), wl_addr_size);
module_manager.add_port(module_id, wl_addr_port, ModuleManager::MODULE_INPUT_PORT); module_manager.add_port(module_id, wl_addr_port, ModuleManager::MODULE_INPUT_PORT);
@ -798,8 +902,8 @@ void add_top_module_sram_ports(ModuleManager& module_manager,
module_manager.add_port(module_id, en_port, ModuleManager::MODULE_INPUT_PORT); module_manager.add_port(module_id, en_port, ModuleManager::MODULE_INPUT_PORT);
size_t max_num_config_bits = 0; size_t max_num_config_bits = 0;
for (const size_t& curr_num_config_bits : num_config_bits) { for (const auto& curr_num_config_bits : num_config_bits) {
max_num_config_bits = std::max(max_num_config_bits, curr_num_config_bits); max_num_config_bits = std::max(max_num_config_bits, curr_num_config_bits.first);
} }
BasicPort addr_port(std::string(DECODER_ADDRESS_PORT_NAME), max_num_config_bits); BasicPort addr_port(std::string(DECODER_ADDRESS_PORT_NAME), max_num_config_bits);
@ -910,7 +1014,7 @@ static
void add_top_module_nets_cmos_memory_bank_config_bus(ModuleManager& module_manager, void add_top_module_nets_cmos_memory_bank_config_bus(ModuleManager& module_manager,
DecoderLibrary& decoder_lib, DecoderLibrary& decoder_lib,
const ModuleId& top_module, const ModuleId& top_module,
const vtr::vector<ConfigRegionId, size_t>& num_config_bits) { const TopModuleNumConfigBits& num_config_bits) {
/* Find Enable port from the top-level module */ /* Find Enable port from the top-level module */
ModulePortId en_port = module_manager.find_module_port(top_module, std::string(DECODER_ENABLE_PORT_NAME)); ModulePortId en_port = module_manager.find_module_port(top_module, std::string(DECODER_ENABLE_PORT_NAME));
BasicPort en_port_info = module_manager.module_port(top_module, en_port); BasicPort en_port_info = module_manager.module_port(top_module, en_port);
@ -935,13 +1039,13 @@ void add_top_module_nets_cmos_memory_bank_config_bus(ModuleManager& module_manag
/* Each memory bank has a unified number of BL/WLs */ /* Each memory bank has a unified number of BL/WLs */
size_t num_bls = 0; size_t num_bls = 0;
for (const size_t& curr_config_bits : num_config_bits) { for (const auto& curr_config_bits : num_config_bits) {
num_bls = std::max(num_bls, find_memory_decoder_data_size(curr_config_bits)); num_bls = std::max(num_bls, find_memory_decoder_data_size(curr_config_bits.first));
} }
size_t num_wls = 0; size_t num_wls = 0;
for (const size_t& curr_config_bits : num_config_bits) { for (const auto& curr_config_bits : num_config_bits) {
num_wls = std::max(num_wls, find_memory_decoder_data_size(curr_config_bits)); num_wls = std::max(num_wls, find_memory_decoder_data_size(curr_config_bits.first));
} }
/* Create separated memory bank circuitry, i.e., BL/WL decoders for each region */ /* Create separated memory bank circuitry, i.e., BL/WL decoders for each region */
@ -1582,7 +1686,7 @@ static
void add_top_module_nets_cmos_memory_frame_config_bus(ModuleManager& module_manager, void add_top_module_nets_cmos_memory_frame_config_bus(ModuleManager& module_manager,
DecoderLibrary& decoder_lib, DecoderLibrary& decoder_lib,
const ModuleId& top_module, const ModuleId& top_module,
const vtr::vector<ConfigRegionId, size_t>& num_config_bits) { const TopModuleNumConfigBits& num_config_bits) {
/* Find the number of address bits for the top-level module */ /* Find the number of address bits for the top-level module */
ModulePortId top_addr_port = module_manager.find_module_port(top_module, std::string(DECODER_ADDRESS_PORT_NAME)); ModulePortId top_addr_port = module_manager.find_module_port(top_module, std::string(DECODER_ADDRESS_PORT_NAME));
BasicPort top_addr_port_info = module_manager.module_port(top_module, top_addr_port); BasicPort top_addr_port_info = module_manager.module_port(top_module, top_addr_port);
@ -1598,7 +1702,7 @@ void add_top_module_nets_cmos_memory_frame_config_bus(ModuleManager& module_mana
* - The number of address bits of the configurable child is the same as top-level * - The number of address bits of the configurable child is the same as top-level
*/ */
if ( (1 == module_manager.region_configurable_children(top_module, config_region).size()) if ( (1 == module_manager.region_configurable_children(top_module, config_region).size())
&& (num_config_bits[config_region] == top_addr_size)) { && (num_config_bits[config_region].first == top_addr_size)) {
add_top_module_nets_cmos_memory_frame_short_config_bus(module_manager, top_module, config_region); add_top_module_nets_cmos_memory_frame_short_config_bus(module_manager, top_module, config_region);
} else { } else {
add_top_module_nets_cmos_memory_frame_decoder_config_bus(module_manager, decoder_lib, top_module, config_region); add_top_module_nets_cmos_memory_frame_decoder_config_bus(module_manager, decoder_lib, top_module, config_region);
@ -1654,8 +1758,10 @@ static
void add_top_module_nets_cmos_memory_config_bus(ModuleManager& module_manager, void add_top_module_nets_cmos_memory_config_bus(ModuleManager& module_manager,
DecoderLibrary& decoder_lib, DecoderLibrary& decoder_lib,
const ModuleId& parent_module, const ModuleId& parent_module,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model,
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const vtr::vector<ConfigRegionId, size_t>& num_config_bits) { const TopModuleNumConfigBits& num_config_bits) {
switch (config_protocol.type()) { switch (config_protocol.type()) {
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
add_module_nets_cmos_flatten_memory_config_bus(module_manager, parent_module, add_module_nets_cmos_flatten_memory_config_bus(module_manager, parent_module,
@ -1670,6 +1776,9 @@ void add_top_module_nets_cmos_memory_config_bus(ModuleManager& module_manager,
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
add_top_module_nets_cmos_memory_bank_config_bus(module_manager, decoder_lib, parent_module, num_config_bits); add_top_module_nets_cmos_memory_bank_config_bus(module_manager, decoder_lib, parent_module, num_config_bits);
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
add_top_module_nets_cmos_ql_memory_bank_config_bus(module_manager, decoder_lib, parent_module, circuit_lib, sram_model, num_config_bits);
break;
case CONFIG_MEM_FRAME_BASED: case CONFIG_MEM_FRAME_BASED:
add_top_module_nets_cmos_memory_frame_config_bus(module_manager, decoder_lib, parent_module, num_config_bits); add_top_module_nets_cmos_memory_frame_config_bus(module_manager, decoder_lib, parent_module, num_config_bits);
break; break;
@ -1715,16 +1824,20 @@ void add_top_module_nets_cmos_memory_config_bus(ModuleManager& module_manager,
void add_top_module_nets_memory_config_bus(ModuleManager& module_manager, void add_top_module_nets_memory_config_bus(ModuleManager& module_manager,
DecoderLibrary& decoder_lib, DecoderLibrary& decoder_lib,
const ModuleId& parent_module, const ModuleId& parent_module,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model,
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const e_circuit_model_design_tech& mem_tech, const e_circuit_model_design_tech& mem_tech,
const vtr::vector<ConfigRegionId, size_t>& num_config_bits) { const TopModuleNumConfigBits& num_config_bits) {
vtr::ScopedStartFinishTimer timer("Add module nets for configuration buses"); vtr::ScopedStartFinishTimer timer("Add module nets for configuration buses");
switch (mem_tech) { switch (mem_tech) {
case CIRCUIT_MODEL_DESIGN_CMOS: case CIRCUIT_MODEL_DESIGN_CMOS:
add_top_module_nets_cmos_memory_config_bus(module_manager, decoder_lib, add_top_module_nets_cmos_memory_config_bus(module_manager, decoder_lib,
parent_module, parent_module,
circuit_lib,
sram_model,
config_protocol, config_protocol,
num_config_bits); num_config_bits);
break; break;

17
openfpga/src/fabric/build_top_module_memory.h
View File

@ -18,6 +18,7 @@
#include "device_rr_gsb.h" #include "device_rr_gsb.h"
#include "fabric_key.h" #include "fabric_key.h"
#include "config_protocol.h" #include "config_protocol.h"
#include "build_top_module_memory_utils.h"
/******************************************************************** /********************************************************************
* Function declaration * Function declaration
@ -48,25 +49,27 @@ int load_top_module_memory_modules_from_fabric_key(ModuleManager& module_manager
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const FabricKey& fabric_key); const FabricKey& fabric_key);
vtr::vector<ConfigRegionId, size_t> find_top_module_regional_num_config_bit(const ModuleManager& module_manager, TopModuleNumConfigBits find_top_module_regional_num_config_bit(const ModuleManager& module_manager,
const ModuleId& top_module, const ModuleId& top_module,
const CircuitLibrary& circuit_lib, const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model, const CircuitModelId& sram_model,
const e_config_protocol_type& config_protocol_type); const e_config_protocol_type& config_protocol_type);
void add_top_module_sram_ports(ModuleManager& module_manager, void add_top_module_sram_ports(ModuleManager& module_manager,
const ModuleId& module_id, const ModuleId& module_id,
const CircuitLibrary& circuit_lib, const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model, const CircuitModelId& sram_model,
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const vtr::vector<ConfigRegionId, size_t>& num_config_bits); const TopModuleNumConfigBits& num_config_bits);
void add_top_module_nets_memory_config_bus(ModuleManager& module_manager, void add_top_module_nets_memory_config_bus(ModuleManager& module_manager,
DecoderLibrary& decoder_lib, DecoderLibrary& decoder_lib,
const ModuleId& parent_module, const ModuleId& parent_module,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model,
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const e_circuit_model_design_tech& mem_tech, const e_circuit_model_design_tech& mem_tech,
const vtr::vector<ConfigRegionId, size_t>& num_config_bits); const TopModuleNumConfigBits& num_config_bits);
} /* end namespace openfpga */ } /* end namespace openfpga */

417
openfpga/src/fabric/build_top_module_memory_bank.cpp Normal file
View File

@ -0,0 +1,417 @@
/********************************************************************
* This file includes functions that are used to organize memories
* in the top module of FPGA fabric
*******************************************************************/
#include <cmath>
#include <limits>
/* Headers from vtrutil library */
#include "vtr_assert.h"
#include "vtr_log.h"
#include "vtr_time.h"
/* Headers from vpr library */
#include "vpr_utils.h"
/* Headers from openfpgashell library */
#include "command_exit_codes.h"
#include "rr_gsb_utils.h"
#include "openfpga_reserved_words.h"
#include "openfpga_naming.h"
#include "memory_utils.h"
#include "decoder_library_utils.h"
#include "module_manager_utils.h"
#include "memory_bank_utils.h"
#include "build_decoder_modules.h"
#include "build_top_module_memory_bank.h"
/* begin namespace openfpga */
namespace openfpga {
/*********************************************************************
* Top-level function to add nets for quicklogic memory banks
* Each configuration region has independent memory bank circuitry
* - Find the number of BLs and WLs required for each region
* - Create BL and WL decoders, and add them to decoder library
* - Create nets to connect from top-level module inputs to inputs of decoders
* - Create nets to connect from outputs of decoders to BL/WL of configurable children
*
* Detailed schematic of how memory banks are connected in the top-level:
* Consider a random Region X, local BL address lines are aligned to the LSB of the
* top-level BL address lines
*
* top_bl_addr[N-1:0]
* ^
* | local_bl_addr[N-1:0]
* |
* +-----+------------------+
* | | |
* | +-------------------+ |
* | | Word Line Decoder | |
* | +-------------------+ |
* | |
*
* The BL/WL decoders should have the same circuit designs no matter what region
* they are placed even when the number of configuration bits are different
* from one region to another!
* This is designed to avoid any address collision between memory banks
* since they are programmed in the same clock cycle
* For example:
* - Memory Bank A has 36 memory cells.
* Its BL decoder has 3 address bit and 6 data output bit
* Its WL decoder has 3 address bit and 6 data output bit
* - Memory Bank B has 16 memory cells.
* Its BL decoder has 2 address bit and 4 data output bit
* Its WL decoder has 2 address bit and 4 data output bit
* - If we try to program the 36th memory cell in bank A
* the BL address will be 3'b110
* the WL address will be 3'b110
* the data input will be 1'b0
* - If we try to program the 4th memory cell in bank A
* the BL address will be 3'b010
* the WL address will be 3'b010
* the data input will be 1'b1
* However, in both cases, this will trigger a parasitic programming in bank B
* the BL address will be 2'b10
* the WL address will be 2'b10
* Assume the data input is expected to be 1'b1 for bank B
* but it will be overwritten to 1'b0 when programming the 36th cell in bank A!
*
* Detailed schematic of each memory bank:
* @note The numbers are just made to show a simplified example, practical cases are more complicated!
*
* WL_enable WL address
* | |
* v v
* +-----------------------------------------------+
* | Word Line Decoder |
* +-----------------------------------------------+
* +---------+ | | |
* BL | | | | |
* enable ---->| |-----------+---------------+---- ... |------+--> BL[0:2]
* | | | | | | | |
* | | | v | v | v
* | Bit | | +-------+ | +-------+ | +------+
* BL | Line | +-->| SRAM | +-->| SRAM | +->| SRAM |
* address ---->| Decoder | | | [0:8] | | | [0:5] | ... | | [0:7]|
* | | | +-------+ | +-------+ | +------+
* | | | | |
* | |-----------+--------------+--------- | -----+--> BL[0:9]
* | | | | | | | |
* | | | v | v | v
* | | | +-------+ | +-------+ | +-------+
* | | +-->| SRAM | | | SRAM | +->| SRAM |
* | | | | [0:80]| | | [0:63]| ... | | [0:31]|
* | | | +-------+ | +-------+ | +-------+
* | | | |
* | | | ... ... ... | ...
* | | | | |
* | |-----------+---------------+---- --- | -----+--> BL[0:3]
* | | | | | | | |
* | | | v | v | v
* | | | +-------+ | +-------+ | +-------+
* | | +-->| SRAM | +-->| SRAM | +->| SRAM |
* | | | |[0:5] | | | [0:8] | ... | | [0:2] |
* | | | +-------+ | +-------+ | +-------+
* BL | | v v v
* data_in ---->| | WL[0:9] WL[0:7] WL[0:4]
* +---------+
*
**********************************************************************/
void add_top_module_nets_cmos_ql_memory_bank_config_bus(ModuleManager& module_manager,
DecoderLibrary& decoder_lib,
const ModuleId& top_module,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model,
const TopModuleNumConfigBits& num_config_bits) {
/* Find Enable port from the top-level module */
ModulePortId en_port = module_manager.find_module_port(top_module, std::string(DECODER_ENABLE_PORT_NAME));
BasicPort en_port_info = module_manager.module_port(top_module, en_port);
/* Find data-in port from the top-level module */
ModulePortId din_port = module_manager.find_module_port(top_module, std::string(DECODER_DATA_IN_PORT_NAME));
BasicPort din_port_info = module_manager.module_port(top_module, din_port);
/* Data in port should match the number of configuration regions */
VTR_ASSERT(din_port_info.get_width() == module_manager.regions(top_module).size());
/* Find BL and WL address port from the top-level module */
ModulePortId bl_addr_port = module_manager.find_module_port(top_module, std::string(DECODER_BL_ADDRESS_PORT_NAME));
BasicPort bl_addr_port_info = module_manager.module_port(top_module, bl_addr_port);
ModulePortId wl_addr_port = module_manager.find_module_port(top_module, std::string(DECODER_WL_ADDRESS_PORT_NAME));
BasicPort wl_addr_port_info = module_manager.module_port(top_module, wl_addr_port);
/* Find the top-level number of BLs and WLs required to access each memory bit */
size_t bl_addr_size = bl_addr_port_info.get_width();
size_t wl_addr_size = wl_addr_port_info.get_width();
/* Each memory bank has a unified number of BL/WLs */
size_t num_bls = 0;
for (const auto& curr_config_bits : num_config_bits) {
num_bls = std::max(num_bls, curr_config_bits.first);
}
size_t num_wls = 0;
for (const auto& curr_config_bits : num_config_bits) {
num_wls = std::max(num_wls, curr_config_bits.second);
}
/* Create separated memory bank circuitry, i.e., BL/WL decoders for each region */
for (const ConfigRegionId& config_region : module_manager.regions(top_module)) {
/**************************************************************
* Add the BL decoder module
* Search the decoder library
* If we find one, we use the module.
* Otherwise, we create one and add it to the decoder library
*/
DecoderId bl_decoder_id = decoder_lib.find_decoder(bl_addr_size, num_bls,
true, true, false);
if (DecoderId::INVALID() == bl_decoder_id) {
bl_decoder_id = decoder_lib.add_decoder(bl_addr_size, num_bls, true, true, false);
}
VTR_ASSERT(DecoderId::INVALID() != bl_decoder_id);
/* Create a module if not existed yet */
std::string bl_decoder_module_name = generate_memory_decoder_with_data_in_subckt_name(bl_addr_size, num_bls);
ModuleId bl_decoder_module = module_manager.find_module(bl_decoder_module_name);
if (ModuleId::INVALID() == bl_decoder_module) {
/* BL decoder has the same ports as the frame-based decoders
* We reuse it here
*/
bl_decoder_module = build_bl_memory_decoder_module(module_manager,
decoder_lib,
bl_decoder_id);
}
VTR_ASSERT(ModuleId::INVALID() != bl_decoder_module);
size_t curr_bl_decoder_instance_id = module_manager.num_instance(top_module, bl_decoder_module);
module_manager.add_child_module(top_module, bl_decoder_module);
/**************************************************************
* Add the WL decoder module
* Search the decoder library
* If we find one, we use the module.
* Otherwise, we create one and add it to the decoder library
*/
DecoderId wl_decoder_id = decoder_lib.find_decoder(wl_addr_size, num_wls,
true, false, false);
if (DecoderId::INVALID() == wl_decoder_id) {
wl_decoder_id = decoder_lib.add_decoder(wl_addr_size, num_wls, true, false, false);
}
VTR_ASSERT(DecoderId::INVALID() != wl_decoder_id);
/* Create a module if not existed yet */
std::string wl_decoder_module_name = generate_memory_decoder_subckt_name(wl_addr_size, num_wls);
ModuleId wl_decoder_module = module_manager.find_module(wl_decoder_module_name);
if (ModuleId::INVALID() == wl_decoder_module) {
/* BL decoder has the same ports as the frame-based decoders
* We reuse it here
*/
wl_decoder_module = build_wl_memory_decoder_module(module_manager,
decoder_lib,
wl_decoder_id);
}
VTR_ASSERT(ModuleId::INVALID() != wl_decoder_module);
size_t curr_wl_decoder_instance_id = module_manager.num_instance(top_module, wl_decoder_module);
module_manager.add_child_module(top_module, wl_decoder_module);
/**************************************************************
* Add module nets from the top module to BL decoder's inputs
*/
ModulePortId bl_decoder_en_port = module_manager.find_module_port(bl_decoder_module, std::string(DECODER_ENABLE_PORT_NAME));
BasicPort bl_decoder_en_port_info = module_manager.module_port(bl_decoder_module, bl_decoder_en_port);
ModulePortId bl_decoder_addr_port = module_manager.find_module_port(bl_decoder_module, std::string(DECODER_ADDRESS_PORT_NAME));
BasicPort bl_decoder_addr_port_info = module_manager.module_port(bl_decoder_module, bl_decoder_addr_port);
ModulePortId bl_decoder_din_port = module_manager.find_module_port(bl_decoder_module, std::string(DECODER_DATA_IN_PORT_NAME));
BasicPort bl_decoder_din_port_info = module_manager.module_port(bl_decoder_module, bl_decoder_din_port);
/* Data in port of the local BL decoder should always be 1 */
VTR_ASSERT(1 == bl_decoder_din_port_info.get_width());
/* Top module Enable port -> BL Decoder Enable port */
add_module_bus_nets(module_manager,
top_module,
top_module, 0, en_port,
bl_decoder_module, curr_bl_decoder_instance_id, bl_decoder_en_port);
/* Top module Address port -> BL Decoder Address port */
add_module_bus_nets(module_manager,
top_module,
top_module, 0, bl_addr_port,
bl_decoder_module, curr_bl_decoder_instance_id, bl_decoder_addr_port);
/* Top module data_in port -> BL Decoder data_in port:
* Note that each region has independent data_in connection from the top-level module
* The pin index is the configuration region index
*/
ModuleNetId din_net = create_module_source_pin_net(module_manager, top_module,
top_module, 0,
din_port,
din_port_info.pins()[size_t(config_region)]);
VTR_ASSERT(ModuleNetId::INVALID() != din_net);
/* Configure the net sink */
module_manager.add_module_net_sink(top_module, din_net, bl_decoder_module, curr_bl_decoder_instance_id, bl_decoder_din_port, bl_decoder_din_port_info.pins()[0]);
/**************************************************************
* Add module nets from the top module to WL decoder's inputs
*/
ModulePortId wl_decoder_en_port = module_manager.find_module_port(wl_decoder_module, std::string(DECODER_ENABLE_PORT_NAME));
BasicPort wl_decoder_en_port_info = module_manager.module_port(wl_decoder_module, wl_decoder_en_port);
ModulePortId wl_decoder_addr_port = module_manager.find_module_port(wl_decoder_module, std::string(DECODER_ADDRESS_PORT_NAME));
BasicPort wl_decoder_addr_port_info = module_manager.module_port(wl_decoder_module, bl_decoder_addr_port);
/* Top module Enable port -> WL Decoder Enable port */
add_module_bus_nets(module_manager,
top_module,
top_module, 0, en_port,
wl_decoder_module, curr_wl_decoder_instance_id, wl_decoder_en_port);
/* Top module Address port -> WL Decoder Address port */
add_module_bus_nets(module_manager,
top_module,
top_module, 0, wl_addr_port,
wl_decoder_module, curr_wl_decoder_instance_id, wl_decoder_addr_port);
/**************************************************************
* Precompute the BLs and WLs distribution across the FPGA fabric
* The distribution is a matrix which contains the starting index of BL/WL for each column or row
*/
std::pair<int, int> child_x_range = compute_memory_bank_regional_configurable_child_x_range(module_manager, top_module, config_region);
std::pair<int, int> child_y_range = compute_memory_bank_regional_configurable_child_y_range(module_manager, top_module, config_region);
std::map<int, size_t> num_bls_per_tile = compute_memory_bank_regional_bitline_numbers_per_tile(module_manager, top_module,
config_region,
circuit_lib, sram_model);
std::map<int, size_t> num_wls_per_tile = compute_memory_bank_regional_wordline_numbers_per_tile(module_manager, top_module,
config_region,
circuit_lib, sram_model,
num_bls_per_tile);
std::map<int, size_t> bl_start_index_per_tile = compute_memory_bank_regional_blwl_start_index_per_tile(child_x_range, num_bls_per_tile);
std::map<int, size_t> wl_start_index_per_tile = compute_memory_bank_regional_blwl_start_index_per_tile(child_y_range, num_wls_per_tile);
/**************************************************************
* Add nets from BL data out to each configurable child
* BL data output pins are connected to the BL input pins of each PB/CB/SB
* For all the PB/CB/SB in the same column, they share the same set of BLs
* A quick example
*
* BL[i .. i + sqrt(N)]
* |
* | CLB[1][H]
* | +---------+
* | | SRAM |
* +-->| [0..N] |
* | +---------+
* |
* ...
* | CLB[1][1]
* | +---------+
* | | SRAM |
* +-->| [0..N] |
* | +---------+
* |
*/
ModulePortId bl_decoder_dout_port = module_manager.find_module_port(bl_decoder_module, std::string(DECODER_DATA_OUT_PORT_NAME));
BasicPort bl_decoder_dout_port_info = module_manager.module_port(bl_decoder_module, bl_decoder_dout_port);
for (size_t child_id = 0; child_id < module_manager.region_configurable_children(top_module, config_region).size(); ++child_id) {
ModuleId child_module = module_manager.region_configurable_children(top_module, config_region)[child_id];
vtr::Point<int> coord = module_manager.region_configurable_child_coordinates(top_module, config_region)[child_id];
int child_num_unique_blwls = num_bls_per_tile.at(coord.x());
size_t child_instance = module_manager.region_configurable_child_instances(top_module, config_region)[child_id];
/* Find the BL port */
ModulePortId child_bl_port = module_manager.find_module_port(child_module, std::string(MEMORY_BL_PORT_NAME));
BasicPort child_bl_port_info = module_manager.module_port(child_module, child_bl_port);
size_t cur_bl_index = 0;
for (const size_t& sink_bl_pin : child_bl_port_info.pins()) {
size_t bl_pin_id = bl_start_index_per_tile[coord.x()] + cur_bl_index % child_num_unique_blwls;
/* Find the BL decoder data index:
* It should be the starting index plus an offset which is the residual when divided by the number of BLs in this tile
*/
VTR_ASSERT(bl_pin_id < bl_decoder_dout_port_info.pins().size());
/* Create net */
ModuleNetId net = create_module_source_pin_net(module_manager, top_module,
bl_decoder_module, curr_bl_decoder_instance_id,
bl_decoder_dout_port,
bl_decoder_dout_port_info.pins()[bl_pin_id]);
VTR_ASSERT(ModuleNetId::INVALID() != net);
/* Add net sink */
module_manager.add_module_net_sink(top_module, net,
child_module, child_instance, child_bl_port, sink_bl_pin);
cur_bl_index++;
}
}
/**************************************************************
* Add nets from WL data out to each configurable child
*/
ModulePortId wl_decoder_dout_port = module_manager.find_module_port(wl_decoder_module, std::string(DECODER_DATA_OUT_PORT_NAME));
BasicPort wl_decoder_dout_port_info = module_manager.module_port(wl_decoder_module, wl_decoder_dout_port);
for (size_t child_id = 0; child_id < module_manager.region_configurable_children(top_module, config_region).size(); ++child_id) {
ModuleId child_module = module_manager.region_configurable_children(top_module, config_region)[child_id];
vtr::Point<int> coord = module_manager.region_configurable_child_coordinates(top_module, config_region)[child_id];
int child_num_unique_blwls = num_bls_per_tile.at(coord.x());
size_t child_instance = module_manager.region_configurable_child_instances(top_module, config_region)[child_id];
/* Find the WL port */
ModulePortId child_wl_port = module_manager.find_module_port(child_module, std::string(MEMORY_WL_PORT_NAME));
BasicPort child_wl_port_info = module_manager.module_port(child_module, child_wl_port);
size_t cur_wl_index = 0;
for (const size_t& sink_wl_pin : child_wl_port_info.pins()) {
size_t wl_pin_id = wl_start_index_per_tile[coord.y()] + std::floor(cur_wl_index / child_num_unique_blwls);
VTR_ASSERT(wl_pin_id < wl_decoder_dout_port_info.pins().size());
/* Create net */
ModuleNetId net = create_module_source_pin_net(module_manager, top_module,
wl_decoder_module, curr_wl_decoder_instance_id,
wl_decoder_dout_port,
wl_decoder_dout_port_info.pins()[wl_pin_id]);
VTR_ASSERT(ModuleNetId::INVALID() != net);
/* Add net sink */
module_manager.add_module_net_sink(top_module, net,
child_module, child_instance, child_wl_port, sink_wl_pin);
cur_wl_index++;
}
}
/**************************************************************
* Add the BL and WL decoders to the end of configurable children list
* Note: this MUST be done after adding all the module nets to other regular configurable children
*/
module_manager.add_configurable_child(top_module, bl_decoder_module, curr_bl_decoder_instance_id);
module_manager.add_configurable_child_to_region(top_module,
config_region,
bl_decoder_module,
curr_bl_decoder_instance_id,
module_manager.configurable_children(top_module).size() - 1);
module_manager.add_configurable_child(top_module, wl_decoder_module, curr_wl_decoder_instance_id);
module_manager.add_configurable_child_to_region(top_module,
config_region,
wl_decoder_module,
curr_wl_decoder_instance_id,
module_manager.configurable_children(top_module).size() - 1);
}
}
} /* end namespace openfpga */

33
openfpga/src/fabric/build_top_module_memory_bank.h Normal file
View File

@ -0,0 +1,33 @@
#ifndef BUILD_TOP_MODULE_MEMORY_BANK_H
#define BUILD_TOP_MODULE_MEMORY_BANK_H
/********************************************************************
* Include header files that are required by function declaration
*******************************************************************/
#include <vector>
#include <map>
#include "vtr_vector.h"
#include "vtr_ndmatrix.h"
#include "module_manager.h"
#include "circuit_library.h"
#include "decoder_library.h"
#include "build_top_module_memory_utils.h"
/********************************************************************
* Function declaration
*******************************************************************/
/* begin namespace openfpga */
namespace openfpga {
void add_top_module_nets_cmos_ql_memory_bank_config_bus(ModuleManager& module_manager,
DecoderLibrary& decoder_lib,
const ModuleId& top_module,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model,
const TopModuleNumConfigBits& num_config_bits);
} /* end namespace openfpga */
#endif

28
openfpga/src/fabric/build_top_module_memory_utils.h Normal file
View File

@ -0,0 +1,28 @@
#ifndef BUILD_TOP_MODULE_MEMORY_UTILS_H
#define BUILD_TOP_MODULE_MEMORY_UTILS_H
/********************************************************************
* Include header files that are required by function declaration
*******************************************************************/
#include <vector>
#include <map>
#include "vtr_vector.h"
/********************************************************************
* Function declaration
*******************************************************************/
/* begin namespace openfpga */
namespace openfpga {
/* A data structure to store the number of configuration bits for each configurable region
* of the top-level module.
* For different configuration protocol, the std::pair<size_t, size_t> represents different data
* See details in each function about how the data is organized
*/
typedef vtr::vector<ConfigRegionId, std::pair<size_t, size_t>> TopModuleNumConfigBits;
} /* end namespace openfpga */
#endif

34
openfpga/src/fabric/module_manager.cpp
View File

@ -84,6 +84,13 @@ std::vector<size_t> ModuleManager::configurable_child_instances(const ModuleId&
return configurable_child_instances_[parent_module]; return configurable_child_instances_[parent_module];
} }
std::vector<vtr::Point<int>> ModuleManager::configurable_child_coordinates(const ModuleId& parent_module) const {
/* Validate the module_id */
VTR_ASSERT(valid_module_id(parent_module));
return configurable_child_coordinates_[parent_module];
}
/* Find the source ids of modules */ /* Find the source ids of modules */
ModuleManager::module_net_src_range ModuleManager::module_net_sources(const ModuleId& module, const ModuleNetId& net) const { ModuleManager::module_net_src_range ModuleManager::module_net_sources(const ModuleId& module, const ModuleNetId& net) const {
/* Validate the module_id */ /* Validate the module_id */
@ -135,6 +142,22 @@ std::vector<size_t> ModuleManager::region_configurable_child_instances(const Mod
return region_config_child_instances; return region_config_child_instances;
} }
std::vector<vtr::Point<int>> ModuleManager::region_configurable_child_coordinates(const ModuleId& parent_module,
const ConfigRegionId& region) const {
/* Validate the module_id */
VTR_ASSERT(valid_module_id(parent_module));
VTR_ASSERT(valid_region_id(parent_module, region));
std::vector<vtr::Point<int>> region_config_child_coordinates;
region_config_child_coordinates.reserve(config_region_children_[parent_module][region].size());
for (const size_t& child_id : config_region_children_[parent_module][region]) {
region_config_child_coordinates.push_back(configurable_child_coordinates_[parent_module][child_id]);
}
return region_config_child_coordinates;
}
/****************************************************************************** /******************************************************************************
* Public Accessors * Public Accessors
******************************************************************************/ ******************************************************************************/
@ -534,6 +557,7 @@ ModuleId ModuleManager::add_module(const std::string& name) {
configurable_children_.emplace_back(); configurable_children_.emplace_back();
configurable_child_instances_.emplace_back(); configurable_child_instances_.emplace_back();
configurable_child_regions_.emplace_back(); configurable_child_regions_.emplace_back();
configurable_child_coordinates_.emplace_back();
config_region_ids_.emplace_back(); config_region_ids_.emplace_back();
config_region_children_.emplace_back(); config_region_children_.emplace_back();
@ -716,7 +740,8 @@ void ModuleManager::set_child_instance_name(const ModuleId& parent_module,
*/ */
void ModuleManager::add_configurable_child(const ModuleId& parent_module, void ModuleManager::add_configurable_child(const ModuleId& parent_module,
const ModuleId& child_module, const ModuleId& child_module,
const size_t& child_instance) { const size_t& child_instance,
const vtr::Point<int> coord) {
/* Validate the id of both parent and child modules */ /* Validate the id of both parent and child modules */
VTR_ASSERT ( valid_module_id(parent_module) ); VTR_ASSERT ( valid_module_id(parent_module) );
VTR_ASSERT ( valid_module_id(child_module) ); VTR_ASSERT ( valid_module_id(child_module) );
@ -726,6 +751,7 @@ void ModuleManager::add_configurable_child(const ModuleId& parent_module,
configurable_children_[parent_module].push_back(child_module); configurable_children_[parent_module].push_back(child_module);
configurable_child_instances_[parent_module].push_back(child_instance); configurable_child_instances_[parent_module].push_back(child_instance);
configurable_child_regions_[parent_module].push_back(ConfigRegionId::INVALID()); configurable_child_regions_[parent_module].push_back(ConfigRegionId::INVALID());
configurable_child_coordinates_[parent_module].push_back(coord);
} }
void ModuleManager::reserve_configurable_child(const ModuleId& parent_module, void ModuleManager::reserve_configurable_child(const ModuleId& parent_module,
@ -738,9 +764,12 @@ void ModuleManager::reserve_configurable_child(const ModuleId& parent_module,
if (num_children > configurable_child_instances_[parent_module].size()) { if (num_children > configurable_child_instances_[parent_module].size()) {
configurable_child_instances_[parent_module].reserve(num_children); configurable_child_instances_[parent_module].reserve(num_children);
} }
if (num_children > configurable_child_instances_[parent_module].size()) { if (num_children > configurable_child_regions_[parent_module].size()) {
configurable_child_regions_[parent_module].reserve(num_children); configurable_child_regions_[parent_module].reserve(num_children);
} }
if (num_children > configurable_child_coordinates_[parent_module].size()) {
configurable_child_coordinates_[parent_module].reserve(num_children);
}
} }
ConfigRegionId ModuleManager::add_config_region(const ModuleId& module) { ConfigRegionId ModuleManager::add_config_region(const ModuleId& module) {
@ -981,6 +1010,7 @@ void ModuleManager::clear_configurable_children(const ModuleId& parent_module) {
configurable_children_[parent_module].clear(); configurable_children_[parent_module].clear();
configurable_child_instances_[parent_module].clear(); configurable_child_instances_[parent_module].clear();
configurable_child_regions_[parent_module].clear(); configurable_child_regions_[parent_module].clear();
configurable_child_coordinates_[parent_module].clear();
} }
void ModuleManager::clear_config_region(const ModuleId& parent_module) { void ModuleManager::clear_config_region(const ModuleId& parent_module) {

17
openfpga/src/fabric/module_manager.h
View File

@ -8,6 +8,7 @@
#include <unordered_map> #include <unordered_map>
#include "vtr_vector.h" #include "vtr_vector.h"
#include "vtr_geometry.h"
#include "module_manager_fwd.h" #include "module_manager_fwd.h"
#include "openfpga_port.h" #include "openfpga_port.h"
@ -148,6 +149,8 @@ class ModuleManager {
std::vector<ModuleId> configurable_children(const ModuleId& parent_module) const; std::vector<ModuleId> configurable_children(const ModuleId& parent_module) const;
/* Find all the instances of configurable child modules under a parent module */ /* Find all the instances of configurable child modules under a parent module */
std::vector<size_t> configurable_child_instances(const ModuleId& parent_module) const; std::vector<size_t> configurable_child_instances(const ModuleId& parent_module) const;
/* Find the coordindate of a configurable child module under a parent module */
std::vector<vtr::Point<int>> configurable_child_coordinates(const ModuleId& parent_module) const;
/* Find the source ids of modules */ /* Find the source ids of modules */
module_net_src_range module_net_sources(const ModuleId& module, const ModuleNetId& net) const; module_net_src_range module_net_sources(const ModuleId& module, const ModuleNetId& net) const;
/* Find the sink ids of modules */ /* Find the sink ids of modules */
@ -161,6 +164,10 @@ class ModuleManager {
/* Find all the instances of configurable child modules under a region of a parent module */ /* Find all the instances of configurable child modules under a region of a parent module */
std::vector<size_t> region_configurable_child_instances(const ModuleId& parent_module, std::vector<size_t> region_configurable_child_instances(const ModuleId& parent_module,
const ConfigRegionId& region) const; const ConfigRegionId& region) const;
/* Find all the coordinates of configurable child modules under a region of a parent module */
std::vector<vtr::Point<int>> region_configurable_child_coordinates(const ModuleId& parent_module,
const ConfigRegionId& region) const;
public: /* Public accessors */ public: /* Public accessors */
size_t num_modules() const; size_t num_modules() const;
@ -252,8 +259,13 @@ class ModuleManager {
void add_child_module(const ModuleId& parent_module, const ModuleId& child_module); void add_child_module(const ModuleId& parent_module, const ModuleId& child_module);
/* Set the instance name of a child module */ /* Set the instance name of a child module */
void set_child_instance_name(const ModuleId& parent_module, const ModuleId& child_module, const size_t& instance_id, const std::string& instance_name); void set_child_instance_name(const ModuleId& parent_module, const ModuleId& child_module, const size_t& instance_id, const std::string& instance_name);
/* Add a configurable child module to module */ /* Add a configurable child module to module
void add_configurable_child(const ModuleId& module, const ModuleId& child_module, const size_t& child_instance); * This function also set the coordinate of a configurable child
* The coordinate is a relative position in each region, which is used to
* idenify BL/WL sharing
* By default, it is an invalid coordinate
*/
void add_configurable_child(const ModuleId& module, const ModuleId& child_module, const size_t& child_instance, const vtr::Point<int> coord = vtr::Point<int>(-1, -1));
/* Reserved a number of configurable children /* Reserved a number of configurable children
* for memory efficiency * for memory efficiency
*/ */
@ -350,6 +362,7 @@ class ModuleManager {
vtr::vector<ModuleId, std::vector<ModuleId>> configurable_children_; /* Child modules with configurable memory bits that this module contain */ vtr::vector<ModuleId, std::vector<ModuleId>> configurable_children_; /* Child modules with configurable memory bits that this module contain */
vtr::vector<ModuleId, std::vector<size_t>> configurable_child_instances_; /* Instances of child modules with configurable memory bits that this module contain */ vtr::vector<ModuleId, std::vector<size_t>> configurable_child_instances_; /* Instances of child modules with configurable memory bits that this module contain */
vtr::vector<ModuleId, std::vector<ConfigRegionId>> configurable_child_regions_; /* Instances of child modules with configurable memory bits that this module contain */ vtr::vector<ModuleId, std::vector<ConfigRegionId>> configurable_child_regions_; /* Instances of child modules with configurable memory bits that this module contain */
vtr::vector<ModuleId, std::vector<vtr::Point<int>>> configurable_child_coordinates_; /* Relative coorindates of child modules with configurable memory bits that this module contain */
/* Configurable regions to group the configurable children /* Configurable regions to group the configurable children
* Note: * Note:

3
openfpga/src/fpga_bitstream/build_device_bitstream.cpp
View File

@ -96,7 +96,8 @@ size_t rec_estimate_device_bitstream_num_bits(const ModuleManager& module_manage
/* Memory configuration protocol will have 2 decoders /* Memory configuration protocol will have 2 decoders
* at the top-level * at the top-level
*/ */
if (CONFIG_MEM_MEMORY_BANK == config_protocol_type) { if (CONFIG_MEM_MEMORY_BANK == config_protocol_type
|| CONFIG_MEM_QL_MEMORY_BANK == config_protocol_type) {
VTR_ASSERT(2 <= curr_region_num_config_child); VTR_ASSERT(2 <= curr_region_num_config_child);
curr_region_num_config_child -= 2; curr_region_num_config_child -= 2;
} }

12
openfpga/src/fpga_bitstream/build_fabric_bitstream.cpp
View File

@ -19,6 +19,7 @@
#include "decoder_library_utils.h" #include "decoder_library_utils.h"
#include "bitstream_manager_utils.h" #include "bitstream_manager_utils.h"
#include "build_fabric_bitstream.h" #include "build_fabric_bitstream.h"
#include "build_fabric_bitstream_memory_bank.h"
/* begin namespace openfpga */ /* begin namespace openfpga */
namespace openfpga { namespace openfpga {
@ -489,6 +490,7 @@ void rec_build_module_fabric_dependent_frame_bitstream(const BitstreamManager& b
*******************************************************************/ *******************************************************************/
static static
void build_module_fabric_dependent_bitstream(const ConfigProtocol& config_protocol, void build_module_fabric_dependent_bitstream(const ConfigProtocol& config_protocol,
const CircuitLibrary& circuit_lib,
const BitstreamManager& bitstream_manager, const BitstreamManager& bitstream_manager,
const ConfigBlockId& top_block, const ConfigBlockId& top_block,
const ModuleManager& module_manager, const ModuleManager& module_manager,
@ -575,6 +577,13 @@ void build_module_fabric_dependent_bitstream(const ConfigProtocol& config_protoc
} }
break; break;
} }
case CONFIG_MEM_QL_MEMORY_BANK: {
build_module_fabric_dependent_bitstream_ql_memory_bank(config_protocol, circuit_lib,
bitstream_manager, top_block,
module_manager, top_module,
fabric_bitstream);
break;
}
case CONFIG_MEM_FRAME_BASED: { case CONFIG_MEM_FRAME_BASED: {
/* Find address port size */ /* Find address port size */
@ -694,6 +703,7 @@ void build_module_fabric_dependent_bitstream(const ConfigProtocol& config_protoc
*******************************************************************/ *******************************************************************/
FabricBitstream build_fabric_dependent_bitstream(const BitstreamManager& bitstream_manager, FabricBitstream build_fabric_dependent_bitstream(const BitstreamManager& bitstream_manager,
const ModuleManager& module_manager, const ModuleManager& module_manager,
const CircuitLibrary& circuit_lib,
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const bool& verbose) { const bool& verbose) {
FabricBitstream fabric_bitstream; FabricBitstream fabric_bitstream;
@ -712,7 +722,7 @@ FabricBitstream build_fabric_dependent_bitstream(const BitstreamManager& bitstre
VTR_ASSERT(0 == top_module_name.compare(bitstream_manager.block_name(top_block[0]))); VTR_ASSERT(0 == top_module_name.compare(bitstream_manager.block_name(top_block[0])));
/* Start build-up formally */ /* Start build-up formally */
build_module_fabric_dependent_bitstream(config_protocol, build_module_fabric_dependent_bitstream(config_protocol, circuit_lib,
bitstream_manager, top_block[0], bitstream_manager, top_block[0],
module_manager, top_module, module_manager, top_module,
fabric_bitstream); fabric_bitstream);

2
openfpga/src/fpga_bitstream/build_fabric_bitstream.h
View File

@ -6,6 +6,7 @@
*******************************************************************/ *******************************************************************/
#include <vector> #include <vector>
#include "config_protocol.h" #include "config_protocol.h"
#include "circuit_library.h"
#include "bitstream_manager.h" #include "bitstream_manager.h"
#include "fabric_bitstream.h" #include "fabric_bitstream.h"
#include "module_manager.h" #include "module_manager.h"
@ -19,6 +20,7 @@ namespace openfpga {
FabricBitstream build_fabric_dependent_bitstream(const BitstreamManager& bitstream_manager, FabricBitstream build_fabric_dependent_bitstream(const BitstreamManager& bitstream_manager,
const ModuleManager& module_manager, const ModuleManager& module_manager,
const CircuitLibrary& circuit_lib,
const ConfigProtocol& config_protocol, const ConfigProtocol& config_protocol,
const bool& verbose); const bool& verbose);

269
openfpga/src/fpga_bitstream/build_fabric_bitstream_memory_bank.cpp Normal file
View File

@ -0,0 +1,269 @@
/********************************************************************
* This file includes functions to build fabric dependent bitstream
* for memory bank configuration protocol
*******************************************************************/
#include <string>
#include <cmath>
#include <algorithm>
/* Headers from vtrutil library */
#include "vtr_assert.h"
#include "vtr_log.h"
#include "vtr_time.h"
/* Headers from openfpgautil library */
#include "openfpga_decode.h"
#include "openfpga_reserved_words.h"
#include "openfpga_naming.h"
#include "decoder_library_utils.h"
#include "bitstream_manager_utils.h"
#include "memory_bank_utils.h"
#include "build_fabric_bitstream_memory_bank.h"
/* begin namespace openfpga */
namespace openfpga {
/********************************************************************
* This function aims to build a bitstream for memory-bank protocol
* It will walk through all the configurable children under a module
* in a recursive way, following a Depth-First Search (DFS) strategy
* For each configuration child, we use its instance name as a key to spot the
* configuration bits in bitstream manager.
* Note that it is guarentee that the instance name in module manager is
* consistent with the block names in bitstream manager
* We use this link to reorganize the bitstream in the sequence of memories as we stored
* in the configurable_children() and configurable_child_instances() of each module of module manager
*
* In such configuration organization, each memory cell has an unique index.
* Using this index, we can infer the address codes for both BL and WL decoders.
* Note that, we must get the number of BLs and WLs before using this function!
*******************************************************************/
static
void rec_build_module_fabric_dependent_ql_memory_bank_regional_bitstream(const BitstreamManager& bitstream_manager,
const ConfigBlockId& parent_block,
const ModuleManager& module_manager,
const ModuleId& top_module,
const ModuleId& parent_module,
const ConfigRegionId& config_region,
const size_t& bl_addr_size,
const size_t& wl_addr_size,
const std::map<int, size_t>& num_bls_per_tile,
const std::map<int, size_t>& bl_start_index_per_tile,
const std::map<int, size_t>& num_wls_per_tile,
const std::map<int, size_t>& wl_start_index_per_tile,
vtr::Point<int>& tile_coord,
std::map<vtr::Point<int>, size_t>& cur_mem_index,
FabricBitstream& fabric_bitstream,
const FabricBitRegionId& fabric_bitstream_region) {
/* Depth-first search: if we have any children in the parent_block,
* we dive to the next level first!
*/
if (0 < bitstream_manager.block_children(parent_block).size()) {
/* For top module:
* - Use regional configurable children
* - we will skip the two decoders at the end of the configurable children list
*/
if (parent_module == top_module) {
std::vector<ModuleId> configurable_children = module_manager.region_configurable_children(parent_module, config_region);
VTR_ASSERT(2 <= configurable_children.size());
size_t num_configurable_children = configurable_children.size() - 2;
/* Early exit if there is no configurable children */
if (0 == num_configurable_children) {
/* Ensure that there should be no configuration bits in the parent block */
VTR_ASSERT(0 == bitstream_manager.block_bits(parent_block).size());
return;
}
for (size_t child_id = 0; child_id < num_configurable_children; ++child_id) {
ModuleId child_module = configurable_children[child_id];
size_t child_instance = module_manager.region_configurable_child_instances(parent_module, config_region)[child_id];
tile_coord = module_manager.region_configurable_child_coordinates(parent_module, config_region)[child_id];
/* Get the instance name and ensure it is not empty */
std::string instance_name = module_manager.instance_name(parent_module, child_module, child_instance);
/* Find the child block that matches the instance name! */
ConfigBlockId child_block = bitstream_manager.find_child_block(parent_block, instance_name);
/* We must have one valid block id! */
VTR_ASSERT(true == bitstream_manager.valid_block_id(child_block));
/* Go recursively */
rec_build_module_fabric_dependent_ql_memory_bank_regional_bitstream(bitstream_manager, child_block,
module_manager, top_module, child_module,
config_region,
bl_addr_size, wl_addr_size,
num_bls_per_tile, bl_start_index_per_tile,
num_wls_per_tile, wl_start_index_per_tile,
tile_coord,
cur_mem_index,
fabric_bitstream,
fabric_bitstream_region);
}
} else {
VTR_ASSERT(parent_module != top_module);
/* For other modules:
* - Use configurable children directly
* - no need to exclude decoders as they are not there
*/
std::vector<ModuleId> configurable_children = module_manager.configurable_children(parent_module);
size_t num_configurable_children = configurable_children.size();
/* Early exit if there is no configurable children */
if (0 == num_configurable_children) {
/* Ensure that there should be no configuration bits in the parent block */
VTR_ASSERT(0 == bitstream_manager.block_bits(parent_block).size());
return;
}
for (size_t child_id = 0; child_id < num_configurable_children; ++child_id) {
ModuleId child_module = configurable_children[child_id];
size_t child_instance = module_manager.configurable_child_instances(parent_module)[child_id];
/* Get the instance name and ensure it is not empty */
std::string instance_name = module_manager.instance_name(parent_module, child_module, child_instance);
/* Find the child block that matches the instance name! */
ConfigBlockId child_block = bitstream_manager.find_child_block(parent_block, instance_name);
/* We must have one valid block id! */
VTR_ASSERT(true == bitstream_manager.valid_block_id(child_block));
/* Go recursively */
rec_build_module_fabric_dependent_ql_memory_bank_regional_bitstream(bitstream_manager, child_block,
module_manager, top_module, child_module,
config_region,
bl_addr_size, wl_addr_size,
num_bls_per_tile, bl_start_index_per_tile,
num_wls_per_tile, wl_start_index_per_tile,
tile_coord,
cur_mem_index,
fabric_bitstream,
fabric_bitstream_region);
}
}
/* Ensure that there should be no configuration bits in the parent block */
VTR_ASSERT(0 == bitstream_manager.block_bits(parent_block).size());
return;
}
/* Note that, reach here, it means that this is a leaf node.
* We add the configuration bits to the fabric_bitstream,
* And then, we can return
*/
for (const ConfigBitId& config_bit : bitstream_manager.block_bits(parent_block)) {
FabricBitId fabric_bit = fabric_bitstream.add_bit(config_bit);
/* Find BL address */
size_t cur_bl_index = bl_start_index_per_tile.at(tile_coord.x()) + cur_mem_index[tile_coord] % num_bls_per_tile.at(tile_coord.x());
std::vector<char> bl_addr_bits_vec = itobin_charvec(cur_bl_index, bl_addr_size);
/* Find WL address */
size_t cur_wl_index = wl_start_index_per_tile.at(tile_coord.y()) + std::floor(cur_mem_index[tile_coord] / num_bls_per_tile.at(tile_coord.x()));
std::vector<char> wl_addr_bits_vec = itobin_charvec(cur_wl_index, wl_addr_size);
/* Set BL address */
fabric_bitstream.set_bit_bl_address(fabric_bit, bl_addr_bits_vec);
/* Set WL address */
fabric_bitstream.set_bit_wl_address(fabric_bit, wl_addr_bits_vec);
/* Set data input */
fabric_bitstream.set_bit_din(fabric_bit, bitstream_manager.bit_value(config_bit));
/* Add the bit to the region */
fabric_bitstream.add_bit_to_region(fabric_bitstream_region, fabric_bit);
/* Increase the memory index */
cur_mem_index[tile_coord]++;
}
}
/********************************************************************
* Main function to build a fabric-dependent bitstream
* by considering the configuration protocol types
*******************************************************************/
void build_module_fabric_dependent_bitstream_ql_memory_bank(const ConfigProtocol& config_protocol,
const CircuitLibrary& circuit_lib,
const BitstreamManager& bitstream_manager,
const ConfigBlockId& top_block,
const ModuleManager& module_manager,
const ModuleId& top_module,
FabricBitstream& fabric_bitstream) {
/* Ensure we are in the correct type of configuration protocol*/
VTR_ASSERT(config_protocol.type() == CONFIG_MEM_QL_MEMORY_BANK);
/* Find global BL address port size */
ModulePortId bl_addr_port = module_manager.find_module_port(top_module, std::string(DECODER_BL_ADDRESS_PORT_NAME));
BasicPort bl_addr_port_info = module_manager.module_port(top_module, bl_addr_port);
/* Find global WL address port size */
ModulePortId wl_addr_port = module_manager.find_module_port(top_module, std::string(DECODER_WL_ADDRESS_PORT_NAME));
BasicPort wl_addr_port_info = module_manager.module_port(top_module, wl_addr_port);
/* Reserve bits before build-up */
fabric_bitstream.set_use_address(true);
fabric_bitstream.set_use_wl_address(true);
fabric_bitstream.set_bl_address_length(bl_addr_port_info.get_width());
fabric_bitstream.set_wl_address_length(wl_addr_port_info.get_width());
fabric_bitstream.reserve_bits(bitstream_manager.num_bits());
/* Build bitstreams by region */
for (const ConfigRegionId& config_region : module_manager.regions(top_module)) {
/* Find port information for local BL and WL decoder in this region */
std::vector<ModuleId> configurable_children = module_manager.region_configurable_children(top_module, config_region);
VTR_ASSERT(2 <= configurable_children.size());
ModuleId bl_decoder_module = configurable_children[configurable_children.size() - 2];
ModuleId wl_decoder_module = configurable_children[configurable_children.size() - 1];
ModulePortId bl_port = module_manager.find_module_port(bl_decoder_module, std::string(DECODER_DATA_OUT_PORT_NAME));
BasicPort bl_port_info = module_manager.module_port(bl_decoder_module, bl_port);
ModulePortId wl_port = module_manager.find_module_port(wl_decoder_module, std::string(DECODER_DATA_OUT_PORT_NAME));
BasicPort wl_port_info = module_manager.module_port(wl_decoder_module, wl_port);
/* Build the bitstream for all the blocks in this region */
FabricBitRegionId fabric_bitstream_region = fabric_bitstream.add_region();
/**************************************************************
* Precompute the BLs and WLs distribution across the FPGA fabric
* The distribution is a matrix which contains the starting index of BL/WL for each column or row
*/
std::pair<int, int> child_x_range = compute_memory_bank_regional_configurable_child_x_range(module_manager, top_module, config_region);
std::pair<int, int> child_y_range = compute_memory_bank_regional_configurable_child_y_range(module_manager, top_module, config_region);
std::map<int, size_t> num_bls_per_tile = compute_memory_bank_regional_bitline_numbers_per_tile(module_manager, top_module,
config_region,
circuit_lib, config_protocol.memory_model());
std::map<int, size_t> num_wls_per_tile = compute_memory_bank_regional_wordline_numbers_per_tile(module_manager, top_module,
config_region,
circuit_lib, config_protocol.memory_model(),
num_bls_per_tile);
std::map<int, size_t> bl_start_index_per_tile = compute_memory_bank_regional_blwl_start_index_per_tile(child_x_range, num_bls_per_tile);
std::map<int, size_t> wl_start_index_per_tile = compute_memory_bank_regional_blwl_start_index_per_tile(child_y_range, num_wls_per_tile);
vtr::Point<int> temp_coord;
std::map<vtr::Point<int>, size_t> cur_mem_index;
rec_build_module_fabric_dependent_ql_memory_bank_regional_bitstream(bitstream_manager, top_block,
module_manager, top_module, top_module,
config_region,
bl_addr_port_info.get_width(),
wl_addr_port_info.get_width(),
num_bls_per_tile, bl_start_index_per_tile,
num_wls_per_tile, wl_start_index_per_tile,
temp_coord,
cur_mem_index,
fabric_bitstream,
fabric_bitstream_region);
}
}
} /* end namespace openfpga */

31
openfpga/src/fpga_bitstream/build_fabric_bitstream_memory_bank.h Normal file
View File

@ -0,0 +1,31 @@
#ifndef BUILD_FABRIC_BITSTREAM_MEMORY_BANK_H
#define BUILD_FABRIC_BITSTREAM_MEMORY_BANK_H
/********************************************************************
* Include header files that are required by function declaration
*******************************************************************/
#include <vector>
#include "config_protocol.h"
#include "circuit_library.h"
#include "bitstream_manager.h"
#include "fabric_bitstream.h"
#include "module_manager.h"
/********************************************************************
* Function declaration
*******************************************************************/
/* begin namespace openfpga */
namespace openfpga {
void build_module_fabric_dependent_bitstream_ql_memory_bank(const ConfigProtocol& config_protocol,
const CircuitLibrary& circuit_lib,
const BitstreamManager& bitstream_manager,
const ConfigBlockId& top_block,
const ModuleManager& module_manager,
const ModuleId& top_module,
FabricBitstream& fabric_bitstream);
} /* end namespace openfpga */
#endif

1
openfpga/src/fpga_bitstream/fast_configuration.cpp
View File

@ -87,6 +87,7 @@ bool find_bit_value_to_skip_for_fast_configuration(const e_config_protocol_type&
} }
break; break;
} }
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
case CONFIG_MEM_FRAME_BASED: { case CONFIG_MEM_FRAME_BASED: {
/* Count how many logic '1' and logic '0' bits we can skip */ /* Count how many logic '1' and logic '0' bits we can skip */

1
openfpga/src/fpga_bitstream/write_text_fabric_bitstream.cpp
View File

@ -306,6 +306,7 @@ int write_fabric_bitstream_to_text_file(const BitstreamManager& bitstream_manage
bitstream_manager, bitstream_manager,
fabric_bitstream); fabric_bitstream);
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
status = write_memory_bank_fabric_bitstream_to_text_file(fp, status = write_memory_bank_fabric_bitstream_to_text_file(fp,
apply_fast_configuration, apply_fast_configuration,

1
openfpga/src/fpga_bitstream/write_xml_fabric_bitstream.cpp
View File

@ -103,6 +103,7 @@ int write_fabric_config_bit_to_xml_file(std::fstream& fp,
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
/* Bit line address */ /* Bit line address */
write_tab_to_file(fp, xml_hierarchy_depth + 1); write_tab_to_file(fp, xml_hierarchy_depth + 1);

48
openfpga/src/fpga_verilog/verilog_top_testbench.cpp
View File

@ -29,46 +29,14 @@
#include "verilog_constants.h" #include "verilog_constants.h"
#include "verilog_writer_utils.h" #include "verilog_writer_utils.h"
#include "verilog_testbench_utils.h" #include "verilog_testbench_utils.h"
#include "verilog_top_testbench_memory_bank.h"
#include "verilog_top_testbench.h" #include "verilog_top_testbench.h"
#include "verilog_top_testbench_constants.h"
/* begin namespace openfpga */ /* begin namespace openfpga */
namespace openfpga { namespace openfpga {
/********************************************************************
* Local variables used only in this file
*******************************************************************/
constexpr char* TOP_TESTBENCH_REFERENCE_INSTANCE_NAME = "REF_DUT";
constexpr char* TOP_TESTBENCH_FPGA_INSTANCE_NAME = "FPGA_DUT";
constexpr char* TOP_TESTBENCH_REFERENCE_OUTPUT_POSTFIX = "_benchmark";
constexpr char* TOP_TESTBENCH_FPGA_OUTPUT_POSTFIX = "_fpga";
constexpr char* TOP_TESTBENCH_CHECKFLAG_PORT_POSTFIX = "_flag";
constexpr char* TOP_TESTBENCH_PROG_TASK_NAME = "prog_cycle_task";
constexpr char* TOP_TESTBENCH_SIM_START_PORT_NAME = "sim_start";
constexpr char* TOP_TESTBENCH_ERROR_COUNTER = "nb_error";
constexpr char* TOP_TB_RESET_PORT_NAME = "greset";
constexpr char* TOP_TB_SET_PORT_NAME = "gset";
constexpr char* TOP_TB_PROG_RESET_PORT_NAME = "prog_reset";
constexpr char* TOP_TB_PROG_SET_PORT_NAME = "prog_set";
constexpr char* TOP_TB_CONFIG_DONE_PORT_NAME = "config_done";
constexpr char* TOP_TB_OP_CLOCK_PORT_NAME = "op_clock";
constexpr char* TOP_TB_OP_CLOCK_PORT_PREFIX = "operating_clk_";
constexpr char* TOP_TB_PROG_CLOCK_PORT_NAME = "prog_clock";
constexpr char* TOP_TB_INOUT_REG_POSTFIX = "_reg";
constexpr char* TOP_TB_CLOCK_REG_POSTFIX = "_reg";
constexpr char* TOP_TB_BITSTREAM_LENGTH_VARIABLE = "BITSTREAM_LENGTH";
constexpr char* TOP_TB_BITSTREAM_WIDTH_VARIABLE = "BITSTREAM_WIDTH";
constexpr char* TOP_TB_BITSTREAM_MEM_REG_NAME = "bit_mem";
constexpr char* TOP_TB_BITSTREAM_INDEX_REG_NAME = "bit_index";
constexpr char* TOP_TB_BITSTREAM_ITERATOR_REG_NAME = "ibit";
constexpr char* TOP_TB_BITSTREAM_SKIP_FLAG_REG_NAME = "skip_bits";
constexpr char* AUTOCHECK_TOP_TESTBENCH_VERILOG_MODULE_POSTFIX = "_autocheck_top_tb";
/******************************************************************** /********************************************************************
* Generate a simulation clock port name * Generate a simulation clock port name
* This function is designed to produce a uniform clock naming for these ports * This function is designed to produce a uniform clock naming for these ports
@ -249,6 +217,7 @@ void print_verilog_top_testbench_config_protocol_port(std::fstream& fp,
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
print_verilog_top_testbench_config_chain_port(fp, module_manager, top_module); print_verilog_top_testbench_config_chain_port(fp, module_manager, top_module);
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
print_verilog_top_testbench_memory_bank_port(fp, module_manager, top_module); print_verilog_top_testbench_memory_bank_port(fp, module_manager, top_module);
break; break;
@ -867,6 +836,7 @@ size_t calculate_num_config_clock_cycles(const e_config_protocol_type& sram_orgz
100. * ((float)num_config_clock_cycles / (float)(1 + regional_bitstream_max_size) - 1.)); 100. * ((float)num_config_clock_cycles / (float)(1 + regional_bitstream_max_size) - 1.));
} }
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
/* For fast configuration, we will skip all the zero data points */ /* For fast configuration, we will skip all the zero data points */
num_config_clock_cycles = 1 + build_memory_bank_fabric_bitstream_by_address(fabric_bitstream).size(); num_config_clock_cycles = 1 + build_memory_bank_fabric_bitstream_by_address(fabric_bitstream).size();
@ -1125,6 +1095,7 @@ void print_verilog_top_testbench_configuration_protocol_stimulus(std::fstream& f
break; break;
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
case CONFIG_MEM_FRAME_BASED: { case CONFIG_MEM_FRAME_BASED: {
ModulePortId en_port_id = module_manager.find_module_port(top_module, ModulePortId en_port_id = module_manager.find_module_port(top_module,
@ -1735,6 +1706,13 @@ void print_verilog_full_testbench_bitstream(std::fstream& fp,
module_manager, top_module, module_manager, top_module,
fabric_bitstream); fabric_bitstream);
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
print_verilog_full_testbench_ql_memory_bank_bitstream(fp, bitstream_file,
fast_configuration,
bit_value_to_skip,
module_manager, top_module,
fabric_bitstream);
break;
case CONFIG_MEM_FRAME_BASED: case CONFIG_MEM_FRAME_BASED:
print_verilog_full_testbench_frame_decoder_bitstream(fp, bitstream_file, print_verilog_full_testbench_frame_decoder_bitstream(fp, bitstream_file,
fast_configuration, fast_configuration,

42
openfpga/src/fpga_verilog/verilog_top_testbench_constants.h Normal file
View File

@ -0,0 +1,42 @@
#ifndef VERILOG_TOP_TESTBENCH_CONSTANTS
#define VERILOG_TOP_TESTBENCH_CONSTANTS
/* begin namespace openfpga */
namespace openfpga {
constexpr char* TOP_TESTBENCH_REFERENCE_INSTANCE_NAME = "REF_DUT";
constexpr char* TOP_TESTBENCH_FPGA_INSTANCE_NAME = "FPGA_DUT";
constexpr char* TOP_TESTBENCH_REFERENCE_OUTPUT_POSTFIX = "_benchmark";
constexpr char* TOP_TESTBENCH_FPGA_OUTPUT_POSTFIX = "_fpga";
constexpr char* TOP_TESTBENCH_CHECKFLAG_PORT_POSTFIX = "_flag";
constexpr char* TOP_TESTBENCH_PROG_TASK_NAME = "prog_cycle_task";
constexpr char* TOP_TESTBENCH_SIM_START_PORT_NAME = "sim_start";
constexpr char* TOP_TESTBENCH_ERROR_COUNTER = "nb_error";
constexpr char* TOP_TB_RESET_PORT_NAME = "greset";
constexpr char* TOP_TB_SET_PORT_NAME = "gset";
constexpr char* TOP_TB_PROG_RESET_PORT_NAME = "prog_reset";
constexpr char* TOP_TB_PROG_SET_PORT_NAME = "prog_set";
constexpr char* TOP_TB_CONFIG_DONE_PORT_NAME = "config_done";
constexpr char* TOP_TB_OP_CLOCK_PORT_NAME = "op_clock";
constexpr char* TOP_TB_OP_CLOCK_PORT_PREFIX = "operating_clk_";
constexpr char* TOP_TB_PROG_CLOCK_PORT_NAME = "prog_clock";
constexpr char* TOP_TB_INOUT_REG_POSTFIX = "_reg";
constexpr char* TOP_TB_CLOCK_REG_POSTFIX = "_reg";
constexpr char* TOP_TB_BITSTREAM_LENGTH_VARIABLE = "BITSTREAM_LENGTH";
constexpr char* TOP_TB_BITSTREAM_WIDTH_VARIABLE = "BITSTREAM_WIDTH";
constexpr char* TOP_TB_BITSTREAM_MEM_REG_NAME = "bit_mem";
constexpr char* TOP_TB_BITSTREAM_INDEX_REG_NAME = "bit_index";
constexpr char* TOP_TB_BITSTREAM_ITERATOR_REG_NAME = "ibit";
constexpr char* TOP_TB_BITSTREAM_SKIP_FLAG_REG_NAME = "skip_bits";
constexpr char* AUTOCHECK_TOP_TESTBENCH_VERILOG_MODULE_POSTFIX = "_autocheck_top_tb";
} /* end namespace openfpga */
#endif

175
openfpga/src/fpga_verilog/verilog_top_testbench_memory_bank.cpp Normal file
View File

@ -0,0 +1,175 @@
/********************************************************************
* This file includes functions that are used to create
* an auto-check top-level testbench for a FPGA fabric
*******************************************************************/
#include <fstream>
#include <iomanip>
#include <algorithm>
/* Headers from vtrutil library */
#include "vtr_log.h"
#include "vtr_assert.h"
#include "vtr_time.h"
/* Headers from openfpgautil library */
#include "openfpga_port.h"
#include "openfpga_digest.h"
#include "bitstream_manager_utils.h"
#include "openfpga_reserved_words.h"
#include "openfpga_naming.h"
#include "simulation_utils.h"
#include "openfpga_atom_netlist_utils.h"
#include "fast_configuration.h"
#include "fabric_bitstream_utils.h"
#include "fabric_global_port_info_utils.h"
#include "verilog_constants.h"
#include "verilog_writer_utils.h"
#include "verilog_testbench_utils.h"
#include "verilog_top_testbench_memory_bank.h"
#include "verilog_top_testbench_constants.h"
/* begin namespace openfpga */
namespace openfpga {
void print_verilog_full_testbench_ql_memory_bank_bitstream(std::fstream& fp,
const std::string& bitstream_file,
const bool& fast_configuration,
const bool& bit_value_to_skip,
const ModuleManager& module_manager,
const ModuleId& top_module,
const FabricBitstream& fabric_bitstream) {
/* Validate the file stream */
valid_file_stream(fp);
/* Reorganize the fabric bitstream by the same address across regions */
MemoryBankFabricBitstream fabric_bits_by_addr = build_memory_bank_fabric_bitstream_by_address(fabric_bitstream);
/* For fast configuration, identify the final bitstream size to be used */
size_t num_bits_to_skip = 0;
if (true == fast_configuration) {
num_bits_to_skip = fabric_bits_by_addr.size() - find_memory_bank_fast_configuration_fabric_bitstream_size(fabric_bitstream, bit_value_to_skip);
}
VTR_ASSERT(num_bits_to_skip < fabric_bits_by_addr.size());
/* Feed address and data input pair one by one
* Note: the first cycle is reserved for programming reset
* We should give dummy values
*/
ModulePortId bl_addr_port_id = module_manager.find_module_port(top_module,
std::string(DECODER_BL_ADDRESS_PORT_NAME));
BasicPort bl_addr_port = module_manager.module_port(top_module, bl_addr_port_id);
std::vector<size_t> initial_bl_addr_values(bl_addr_port.get_width(), 0);
ModulePortId wl_addr_port_id = module_manager.find_module_port(top_module,
std::string(DECODER_WL_ADDRESS_PORT_NAME));
BasicPort wl_addr_port = module_manager.module_port(top_module, wl_addr_port_id);
std::vector<size_t> initial_wl_addr_values(wl_addr_port.get_width(), 0);
ModulePortId din_port_id = module_manager.find_module_port(top_module,
std::string(DECODER_DATA_IN_PORT_NAME));
BasicPort din_port = module_manager.module_port(top_module, din_port_id);
std::vector<size_t> initial_din_values(din_port.get_width(), 0);
/* Define a constant for the bitstream length */
print_verilog_define_flag(fp, std::string(TOP_TB_BITSTREAM_LENGTH_VARIABLE), fabric_bits_by_addr.size() - num_bits_to_skip);
print_verilog_define_flag(fp, std::string(TOP_TB_BITSTREAM_WIDTH_VARIABLE), bl_addr_port.get_width() + wl_addr_port.get_width() + din_port.get_width());
/* Declare local variables for bitstream loading in Verilog */
print_verilog_comment(fp, "----- Virtual memory to store the bitstream from external file -----");
fp << "reg [0:`" << TOP_TB_BITSTREAM_WIDTH_VARIABLE << " - 1] ";
fp << TOP_TB_BITSTREAM_MEM_REG_NAME << "[0:`" << TOP_TB_BITSTREAM_LENGTH_VARIABLE << " - 1];";
fp << std::endl;
fp << "reg [$clog2(`" << TOP_TB_BITSTREAM_LENGTH_VARIABLE << "):0] " << TOP_TB_BITSTREAM_INDEX_REG_NAME << ";" << std::endl;
print_verilog_comment(fp, "----- Preload bitstream file to a virtual memory -----");
fp << "initial begin" << std::endl;
fp << "\t";
fp << "$readmemb(\"" << bitstream_file << "\", " << TOP_TB_BITSTREAM_MEM_REG_NAME << ");";
fp << std::endl;
print_verilog_comment(fp, "----- Bit-Line Address port default input -----");
fp << "\t";
fp << generate_verilog_port_constant_values(bl_addr_port, initial_bl_addr_values);
fp << ";";
fp << std::endl;
print_verilog_comment(fp, "----- Word-Line Address port default input -----");
fp << "\t";
fp << generate_verilog_port_constant_values(wl_addr_port, initial_wl_addr_values);
fp << ";";
fp << std::endl;
print_verilog_comment(fp, "----- Data-input port default input -----");
fp << "\t";
fp << generate_verilog_port_constant_values(din_port, initial_din_values);
fp << ";";
fp << std::endl;
fp << "\t";
fp << TOP_TB_BITSTREAM_INDEX_REG_NAME << " <= 0";
fp << ";";
fp << std::endl;
fp << "end";
fp << std::endl;
print_verilog_comment(fp, "----- Begin bitstream loading during configuration phase -----");
BasicPort prog_clock_port(std::string(TOP_TB_PROG_CLOCK_PORT_NAME) + std::string(TOP_TB_CLOCK_REG_POSTFIX), 1);
fp << "always";
fp << " @(negedge " << generate_verilog_port(VERILOG_PORT_CONKT, prog_clock_port) << ")";
fp << " begin";
fp << std::endl;
fp << "\t";
fp << "if (";
fp << TOP_TB_BITSTREAM_INDEX_REG_NAME;
fp << " >= ";
fp << "`" << TOP_TB_BITSTREAM_LENGTH_VARIABLE;
fp << ") begin";
fp << std::endl;
BasicPort config_done_port(std::string(TOP_TB_CONFIG_DONE_PORT_NAME), 1);
fp << "\t\t";
std::vector<size_t> config_done_final_values(config_done_port.get_width(), 1);
fp << generate_verilog_port_constant_values(config_done_port, config_done_final_values, true);
fp << ";" << std::endl;
fp << "\t";
fp << "end else begin";
fp << std::endl;
fp << "\t\t";
fp << "{";
fp << generate_verilog_port(VERILOG_PORT_CONKT, bl_addr_port);
fp << ", ";
fp << generate_verilog_port(VERILOG_PORT_CONKT, wl_addr_port);
fp << ", ";
fp << generate_verilog_port(VERILOG_PORT_CONKT, din_port);
fp << "}";
fp << " <= ";
fp << TOP_TB_BITSTREAM_MEM_REG_NAME << "[" << TOP_TB_BITSTREAM_INDEX_REG_NAME << "]";
fp << ";" << std::endl;
fp << "\t\t";
fp << TOP_TB_BITSTREAM_INDEX_REG_NAME;
fp << " <= ";
fp << TOP_TB_BITSTREAM_INDEX_REG_NAME << " + 1";
fp << ";" << std::endl;
fp << "\t";
fp << "end";
fp << std::endl;
fp << "end";
fp << std::endl;
print_verilog_comment(fp, "----- End bitstream loading during configuration phase -----");
}
} /* end namespace openfpga */

43
openfpga/src/fpga_verilog/verilog_top_testbench_memory_bank.h Normal file
View File

@ -0,0 +1,43 @@
#ifndef VERILOG_TOP_TESTBENCH_MEMORY_BANK
#define VERILOG_TOP_TESTBENCH_MEMORY_BANK
/********************************************************************
* Include header files that are required by function declaration
*******************************************************************/
#include <string>
#include <vector>
#include "module_manager.h"
#include "bitstream_manager.h"
#include "fabric_bitstream.h"
#include "circuit_library.h"
#include "config_protocol.h"
#include "vpr_context.h"
#include "pin_constraints.h"
#include "io_location_map.h"
#include "fabric_global_port_info.h"
#include "vpr_netlist_annotation.h"
#include "simulation_setting.h"
#include "verilog_testbench_options.h"
/********************************************************************
* Function declaration
*******************************************************************/
/* begin namespace openfpga */
namespace openfpga {
/**
* @brief Print stimulus for a FPGA fabric with a memory bank configuration protocol
* where configuration bits are programming in serial (one by one)
*/
void print_verilog_full_testbench_ql_memory_bank_bitstream(std::fstream& fp,
const std::string& bitstream_file,
const bool& fast_configuration,
const bool& bit_value_to_skip,
const ModuleManager& module_manager,
const ModuleId& top_module,
const FabricBitstream& fabric_bitstream);
} /* end namespace openfpga */
#endif

3
openfpga/src/fpga_verilog/verilog_writer_utils.cpp
View File

@ -1010,6 +1010,7 @@ void print_verilog_local_sram_wires(std::fstream& fp,
print_verilog_wire_connection(fp, ccff_tail_local_port, ccff_tail_port, false); print_verilog_wire_connection(fp, ccff_tail_local_port, ccff_tail_port, false);
break; break;
} }
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
/* Generate the name of local wire for the SRAM output and inverted output */ /* Generate the name of local wire for the SRAM output and inverted output */
std::vector<BasicPort> sram_ports; std::vector<BasicPort> sram_ports;
@ -1100,6 +1101,7 @@ void print_verilog_local_config_bus(std::fstream& fp,
*/ */
break; break;
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
/* Two configuration buses should be outputted /* Two configuration buses should be outputted
* One for the regular SRAM ports of a routing multiplexer * One for the regular SRAM ports of a routing multiplexer
@ -1173,6 +1175,7 @@ void print_verilog_rram_mux_config_bus(std::fstream& fp,
*/ */
break; break;
} }
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
/* This is currently most used in ReRAM FPGAs */ /* This is currently most used in ReRAM FPGAs */
/* Print configuration bus to group reserved BL/WLs */ /* Print configuration bus to group reserved BL/WLs */

3
openfpga/src/utils/circuit_library_utils.cpp
View File

@ -94,6 +94,7 @@ size_t find_rram_circuit_num_shared_config_bits(const CircuitLibrary& circuit_li
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
/* Find BL/WL ports */ /* Find BL/WL ports */
std::vector<CircuitPortId> blb_ports = circuit_lib.model_ports_by_type(rram_model, CIRCUIT_MODEL_PORT_BLB); std::vector<CircuitPortId> blb_ports = circuit_lib.model_ports_by_type(rram_model, CIRCUIT_MODEL_PORT_BLB);
@ -175,6 +176,7 @@ size_t find_circuit_num_config_bits(const e_config_protocol_type& config_protoco
switch (config_protocol_type) { switch (config_protocol_type) {
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
break; break;
} }
@ -293,6 +295,7 @@ bool check_configurable_memory_circuit_model(const e_config_protocol_type& confi
break; break;
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_FRAME_BASED: case CONFIG_MEM_FRAME_BASED:
num_err = check_sram_circuit_model_ports(circuit_lib, num_err = check_sram_circuit_model_ports(circuit_lib,
config_mem_circuit_model, config_mem_circuit_model,

18
openfpga/src/utils/decoder_library_utils.cpp
View File

@ -86,6 +86,24 @@ size_t find_memory_decoder_data_size(const size_t& num_mems) {
return (size_t)std::ceil(std::sqrt((float)num_mems)); return (size_t)std::ceil(std::sqrt((float)num_mems));
} }
/***************************************************************************************
* Find the size of WL data lines for a memory decoder to access a memory array
* This function is applicable to a memory bank organization where BL data lines
* is the dominant factor. It means that the BL data lines is strictly an integeter close
* to the square root of the number of memory cells.
* For example, 203 memory cells leads to 15 BLs to control
* The WL data lines may not be exactly the same as the number of BLs.
* Considering the example of 203 memory cells again, when 15 BLs are used, we just need
* 203 / 15 = 13.5555 -> 14 WLs
***************************************************************************************/
size_t find_memory_wl_decoder_data_size(const size_t& num_mems, const size_t& num_bls) {
/* Handle exception: zero BLs should have zero WLs */
if (0 == num_bls) {
return 0;
}
return std::ceil((float)num_mems / (float)num_bls);
}
/*************************************************************************************** /***************************************************************************************
* Try to find if the decoder already exists in the library, * Try to find if the decoder already exists in the library,
* If there is no such decoder, add it to the library * If there is no such decoder, add it to the library

2
openfpga/src/utils/decoder_library_utils.h
View File

@ -17,6 +17,8 @@ size_t find_memory_decoder_addr_size(const size_t& num_mems);
size_t find_memory_decoder_data_size(const size_t& num_mems); size_t find_memory_decoder_data_size(const size_t& num_mems);
size_t find_memory_wl_decoder_data_size(const size_t& num_mems, const size_t& num_bls);
DecoderId add_mux_local_decoder_to_library(DecoderLibrary& decoder_lib, DecoderId add_mux_local_decoder_to_library(DecoderLibrary& decoder_lib,
const size_t data_size); const size_t data_size);

106
openfpga/src/utils/memory_bank_utils.cpp Normal file
View File

@ -0,0 +1,106 @@
/********************************************************************
* This file includes functions that are used to organize memories
* in the top module of FPGA fabric
*******************************************************************/
#include <cmath>
#include <limits>
/* Headers from vtrutil library */
#include "vtr_assert.h"
#include "vtr_log.h"
#include "vtr_time.h"
/* Headers from vpr library */
#include "vpr_utils.h"
/* Headers from openfpgashell library */
#include "command_exit_codes.h"
#include "rr_gsb_utils.h"
#include "openfpga_reserved_words.h"
#include "openfpga_naming.h"
#include "memory_utils.h"
#include "decoder_library_utils.h"
#include "module_manager_utils.h"
#include "memory_bank_utils.h"
/* begin namespace openfpga */
namespace openfpga {
std::pair<int, int> compute_memory_bank_regional_configurable_child_x_range(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region) {
std::pair<int, int> child_x_range(std::numeric_limits<int>::max(), std::numeric_limits<int>::min()); // Deposit an invalid range first: LSB->max(); MSB->min()
for (size_t child_id = 0; child_id < module_manager.region_configurable_children(top_module, config_region).size(); ++child_id) {
vtr::Point<int> coord = module_manager.region_configurable_child_coordinates(top_module, config_region)[child_id];
child_x_range.first = std::min(coord.x(), child_x_range.first);
child_x_range.second = std::max(coord.x(), child_x_range.second);
}
VTR_ASSERT(child_x_range.first <= child_x_range.second);
return child_x_range;
}
std::pair<int, int> compute_memory_bank_regional_configurable_child_y_range(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region) {
std::pair<int, int> child_y_range(std::numeric_limits<int>::max(), std::numeric_limits<int>::min()); // Deposit an invalid range first: LSB->max(); MSB->min()
for (size_t child_id = 0; child_id < module_manager.region_configurable_children(top_module, config_region).size(); ++child_id) {
vtr::Point<int> coord = module_manager.region_configurable_child_coordinates(top_module, config_region)[child_id];
child_y_range.first = std::min(coord.y(), child_y_range.first);
child_y_range.second = std::max(coord.y(), child_y_range.second);
}
VTR_ASSERT(child_y_range.first <= child_y_range.second);
return child_y_range;
}
std::map<int, size_t> compute_memory_bank_regional_bitline_numbers_per_tile(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model) {
std::map<int, size_t> num_bls_per_tile;
for (size_t child_id = 0; child_id < module_manager.region_configurable_children(top_module, config_region).size(); ++child_id) {
ModuleId child_module = module_manager.region_configurable_children(top_module, config_region)[child_id];
vtr::Point<int> coord = module_manager.region_configurable_child_coordinates(top_module, config_region)[child_id];
num_bls_per_tile[coord.x()] = std::max(num_bls_per_tile[coord.x()], find_memory_decoder_data_size(find_module_num_config_bits(module_manager, child_module, circuit_lib, sram_model, CONFIG_MEM_QL_MEMORY_BANK)));
}
return num_bls_per_tile;
}
std::map<int, size_t> compute_memory_bank_regional_wordline_numbers_per_tile(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model,
const std::map<int, size_t>& num_bls_per_tile) {
std::map<int, size_t> num_wls_per_tile;
for (size_t child_id = 0; child_id < module_manager.region_configurable_children(top_module, config_region).size(); ++child_id) {
ModuleId child_module = module_manager.region_configurable_children(top_module, config_region)[child_id];
vtr::Point<int> coord = module_manager.region_configurable_child_coordinates(top_module, config_region)[child_id];
num_wls_per_tile[coord.y()] = std::max(num_wls_per_tile[coord.y()], find_memory_wl_decoder_data_size(find_module_num_config_bits(module_manager, child_module, circuit_lib, sram_model, CONFIG_MEM_QL_MEMORY_BANK), num_bls_per_tile.at(coord.x())));
}
return num_wls_per_tile;
}
std::map<int, size_t> compute_memory_bank_regional_blwl_start_index_per_tile(const std::pair<int, int>& child_xy_range,
const std::map<int, size_t>& num_blwls_per_tile) {
std::map<int, size_t> blwl_start_index_per_tile;
for (int iblwl = child_xy_range.first; iblwl <= child_xy_range.second; ++iblwl) {
if (iblwl == child_xy_range.first) {
blwl_start_index_per_tile[iblwl] = 0;
} else {
blwl_start_index_per_tile[iblwl] = num_blwls_per_tile.at(iblwl - 1) + blwl_start_index_per_tile[iblwl - 1];
}
}
return blwl_start_index_per_tile;
}
} /* end namespace openfpga */

81
openfpga/src/utils/memory_bank_utils.h Normal file
View File

@ -0,0 +1,81 @@
#ifndef MEMORY_BANK_UTILS_H
#define MEMORY_BANK_UTILS_H
/********************************************************************
* Include header files that are required by function declaration
*******************************************************************/
#include <vector>
#include <map>
#include "vtr_vector.h"
#include "vtr_ndmatrix.h"
#include "module_manager.h"
#include "circuit_library.h"
#include "decoder_library.h"
#include "build_top_module_memory_utils.h"
/********************************************************************
* Function declaration
*******************************************************************/
/* begin namespace openfpga */
namespace openfpga {
/**
* @brief Precompute the range of x coordinates of all the configurable children under a specific configuration region
* The lower bound is stored in the first element of the return struct
* The upper bound is stored in the second element of the return struct
*/
std::pair<int, int> compute_memory_bank_regional_configurable_child_x_range(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region);
/**
* @brief Precompute the range of y coordinates of all the configurable children under a specific configuration region
* The lower bound is stored in the first element of the return struct
* The upper bound is stored in the second element of the return struct
*/
std::pair<int, int> compute_memory_bank_regional_configurable_child_y_range(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region);
/**
* @brief Precompute the number of bit lines required by each tile under a specific configuration region
* @note
* Not every index in the range computed by the compute_memory_bank_regional_configurable_child_x_range() function has a postive number of bit lines
* If an empty entry is found (e.g., std::map::find(x) is empty), it means there are not bit lines required in that tile
*/
std::map<int, size_t> compute_memory_bank_regional_bitline_numbers_per_tile(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model);
/**
* @brief Precompute the number of word lines required by each tile under a specific configuration region
* @note
* Not every index in the range computed by the compute_memory_bank_regional_configurable_child_y_range() function has a postive number of word lines
* If an empty entry is found (e.g., std::map::find(y) is empty), it means there are not word lines required in that tile
* @note
* This function requires an input argument which describes number of bitlines per tile. Base on the information, the number of word lines are inferred
* by total number of memores / number of bit lines at a given tile location
* This strategy is chosen because in each column, the number of bit lines are bounded by the tile which consumes most configuation bits. It may reduces
* the use of word lines. For example, a tile[0][0] has only 8 bits, from which we may infer 3 BLs and 3 WLs. However, when tile[0][1] contains 100 bits,
* which will force the number of BLs to be 10. In such case, tile[0][0] only requires 1 WL
*/
std::map<int, size_t> compute_memory_bank_regional_wordline_numbers_per_tile(const ModuleManager& module_manager,
const ModuleId& top_module,
const ConfigRegionId& config_region,
const CircuitLibrary& circuit_lib,
const CircuitModelId& sram_model,
const std::map<int, size_t>& num_bls_per_tile);
/**
* @brief Precompute the BLs and WLs distribution across the FPGA fabric
* The distribution is a matrix which contains the starting index of BL/WL for each column or row
*/
std::map<int, size_t> compute_memory_bank_regional_blwl_start_index_per_tile(const std::pair<int, int>& child_xy_range,
const std::map<int, size_t>& num_blwls_per_tile);
} /* end namespace openfpga */
#endif

7
openfpga/src/utils/memory_utils.cpp
View File

@ -74,6 +74,7 @@ std::map<std::string, BasicPort> generate_cmos_mem_module_port2port_map(const Ba
port2port_name_map[generate_configurable_memory_inverted_data_out_name()] = mem_output_bus_ports[1]; port2port_name_map[generate_configurable_memory_inverted_data_out_name()] = mem_output_bus_ports[1];
break; break;
} }
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
/* TODO: */ /* TODO: */
break; break;
@ -131,6 +132,7 @@ std::map<std::string, BasicPort> generate_rram_mem_module_port2port_map(const Ba
port2port_name_map[generate_configurable_memory_inverted_data_out_name()] = mem_output_bus_ports[1]; port2port_name_map[generate_configurable_memory_inverted_data_out_name()] = mem_output_bus_ports[1];
break; break;
} }
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
/* TODO: link BL/WL/Reserved Ports to the inputs of a memory module */ /* TODO: link BL/WL/Reserved Ports to the inputs of a memory module */
break; break;
@ -189,6 +191,7 @@ void update_cmos_mem_module_config_bus(const e_config_protocol_type& sram_orgz_t
*/ */
VTR_ASSERT(true == config_bus.rotate(1)); VTR_ASSERT(true == config_bus.rotate(1));
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
/* In this case, a memory module has a number of BL/WL and BLB/WLB (possibly). /* In this case, a memory module has a number of BL/WL and BLB/WLB (possibly).
* LSB and MSB of configuration bus will be shifted by the number of BL/WL/BLB/WLB. * LSB and MSB of configuration bus will be shifted by the number of BL/WL/BLB/WLB.
@ -219,6 +222,7 @@ void update_rram_mem_module_config_bus(const e_config_protocol_type& sram_orgz_t
*/ */
VTR_ASSERT(true == config_bus.rotate(1)); VTR_ASSERT(true == config_bus.rotate(1));
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
/* In this case, a memory module contains unique BL/WL or BLB/WLB, /* In this case, a memory module contains unique BL/WL or BLB/WLB,
* which are not shared with other modules * which are not shared with other modules
@ -275,6 +279,7 @@ bool check_mem_config_bus(const e_config_protocol_type& sram_orgz_type,
*/ */
return (local_expected_msb == config_bus.get_msb()); return (local_expected_msb == config_bus.get_msb());
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
/* TODO: comment on why /* TODO: comment on why
*/ */
@ -319,6 +324,7 @@ std::vector<std::string> generate_sram_port_names(const CircuitLibrary& circuit_
model_port_types.push_back(CIRCUIT_MODEL_PORT_OUTPUT); model_port_types.push_back(CIRCUIT_MODEL_PORT_OUTPUT);
break; break;
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
std::vector<e_circuit_model_port_type> ports_to_search; std::vector<e_circuit_model_port_type> ports_to_search;
ports_to_search.push_back(CIRCUIT_MODEL_PORT_BL); ports_to_search.push_back(CIRCUIT_MODEL_PORT_BL);
@ -373,6 +379,7 @@ size_t generate_sram_port_size(const e_config_protocol_type sram_orgz_type,
/* CCFF head/tail are single-bit ports */ /* CCFF head/tail are single-bit ports */
sram_port_size = 1; sram_port_size = 1;
break; break;
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
break; break;
case CONFIG_MEM_FRAME_BASED: case CONFIG_MEM_FRAME_BASED:

3
openfpga/src/utils/module_manager_utils.cpp
View File

@ -313,6 +313,7 @@ void add_sram_ports_to_module_manager(ModuleManager& module_manager,
/* Add ports to the module manager */ /* Add ports to the module manager */
switch (sram_orgz_type) { switch (sram_orgz_type) {
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
for (const std::string& sram_port_name : sram_port_names) { for (const std::string& sram_port_name : sram_port_names) {
/* Add generated ports to the ModuleManager */ /* Add generated ports to the ModuleManager */
@ -1288,6 +1289,7 @@ void add_module_nets_cmos_memory_config_bus(ModuleManager& module_manager,
break; break;
} }
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
add_module_nets_cmos_flatten_memory_config_bus(module_manager, parent_module, add_module_nets_cmos_flatten_memory_config_bus(module_manager, parent_module,
sram_orgz_type, CIRCUIT_MODEL_PORT_BL); sram_orgz_type, CIRCUIT_MODEL_PORT_BL);
@ -1742,6 +1744,7 @@ size_t find_module_num_config_bits_from_child_modules(ModuleManager& module_mana
switch (sram_orgz_type) { switch (sram_orgz_type) {
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_MEMORY_BANK: {
/* For scan-chain, standalone and memory bank configuration protocol /* For scan-chain, standalone and memory bank configuration protocol
* The number of configuration bits is the sum of configuration bits * The number of configuration bits is the sum of configuration bits

5
openfpga/src/utils/mux_utils.cpp
View File

@ -257,6 +257,7 @@ size_t find_cmos_mux_num_config_bits(const CircuitLibrary& circuit_lib,
size_t num_config_bits = 0; size_t num_config_bits = 0;
switch (sram_orgz_type) { switch (sram_orgz_type) {
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
@ -298,6 +299,7 @@ size_t find_rram_mux_num_config_bits(const CircuitLibrary& circuit_lib,
const e_config_protocol_type& sram_orgz_type) { const e_config_protocol_type& sram_orgz_type) {
size_t num_config_bits = 0; size_t num_config_bits = 0;
switch (sram_orgz_type) { switch (sram_orgz_type) {
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
/* In memory bank, by intensively share the Bit/Word Lines, /* In memory bank, by intensively share the Bit/Word Lines,
* we only need 1 additional BL and WL for each MUX level. * we only need 1 additional BL and WL for each MUX level.
@ -365,6 +367,7 @@ size_t find_cmos_mux_num_shared_config_bits(const e_config_protocol_type& sram_o
size_t num_shared_config_bits = 0; size_t num_shared_config_bits = 0;
switch (sram_orgz_type) { switch (sram_orgz_type) {
case CONFIG_MEM_QL_MEMORY_BANK:
case CONFIG_MEM_MEMORY_BANK: case CONFIG_MEM_MEMORY_BANK:
case CONFIG_MEM_SCAN_CHAIN: case CONFIG_MEM_SCAN_CHAIN:
case CONFIG_MEM_STANDALONE: case CONFIG_MEM_STANDALONE:
@ -388,7 +391,7 @@ size_t find_rram_mux_num_shared_config_bits(const CircuitLibrary& circuit_lib,
const e_config_protocol_type& sram_orgz_type) { const e_config_protocol_type& sram_orgz_type) {
size_t num_shared_config_bits = 0; size_t num_shared_config_bits = 0;
switch (sram_orgz_type) { switch (sram_orgz_type) {
case CONFIG_MEM_MEMORY_BANK: { case CONFIG_MEM_QL_MEMORY_BANK: {
/* In memory bank, the number of shared configuration bits is /* In memory bank, the number of shared configuration bits is
* the sum of largest branch size at each level * the sum of largest branch size at each level
*/ */

198
openfpga_flow/openfpga_arch/k4_N4_40nm_qlbank_openfpga.xml Normal file
View File

@ -0,0 +1,198 @@
<!-- Architecture annotation for OpenFPGA framework
This annotation supports the k6_N10_40nm.xml
- General purpose logic block
- K = 6, N = 10, I = 40
- Single mode
- Routing architecture
- L = 4, fc_in = 0.15, fc_out = 0.1
-->
<openfpga_architecture>
<technology_library>
<device_library>
<device_model name="logic" type="transistor">
<lib type="industry" corner="TOP_TT" ref="M" path="${OPENFPGA_PATH}/openfpga_flow/tech/PTM_45nm/45nm.pm"/>
<design vdd="0.9" pn_ratio="2"/>
<pmos name="pch" chan_length="40e-9" min_width="140e-9" variation="logic_transistor_var"/>
<nmos name="nch" chan_length="40e-9" min_width="140e-9" variation="logic_transistor_var"/>
</device_model>
<device_model name="io" type="transistor">
<lib type="academia" ref="M" path="${OPENFPGA_PATH}/openfpga_flow/tech/PTM_45nm/45nm.pm"/>
<design vdd="2.5" pn_ratio="3"/>
<pmos name="pch_25" chan_length="270e-9" min_width="320e-9" variation="io_transistor_var"/>
<nmos name="nch_25" chan_length="270e-9" min_width="320e-9" variation="io_transistor_var"/>
</device_model>
</device_library>
<variation_library>
<variation name="logic_transistor_var" abs_deviation="0.1" num_sigma="3"/>
<variation name="io_transistor_var" abs_deviation="0.1" num_sigma="3"/>
</variation_library>
</technology_library>
<circuit_library>
<circuit_model type="inv_buf" name="INVTX1" prefix="INVTX1" is_default="true">
<design_technology type="cmos" topology="inverter" size="1"/>
<device_technology device_model_name="logic"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<delay_matrix type="rise" in_port="in" out_port="out">
10e-12
</delay_matrix>
<delay_matrix type="fall" in_port="in" out_port="out">
10e-12
</delay_matrix>
</circuit_model>
<circuit_model type="inv_buf" name="buf4" prefix="buf4" is_default="false">
<design_technology type="cmos" topology="buffer" size="1" num_level="2" f_per_stage="4"/>
<device_technology device_model_name="logic"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<delay_matrix type="rise" in_port="in" out_port="out">
10e-12
</delay_matrix>
<delay_matrix type="fall" in_port="in" out_port="out">
10e-12
</delay_matrix>
</circuit_model>
<circuit_model type="inv_buf" name="tap_buf4" prefix="tap_buf4" is_default="false">
<design_technology type="cmos" topology="buffer" size="1" num_level="3" f_per_stage="4"/>
<device_technology device_model_name="logic"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<delay_matrix type="rise" in_port="in" out_port="out">
10e-12
</delay_matrix>
<delay_matrix type="fall" in_port="in" out_port="out">
10e-12
</delay_matrix>
</circuit_model>
<circuit_model type="pass_gate" name="TGATE" prefix="TGATE" is_default="true">
<design_technology type="cmos" topology="transmission_gate" nmos_size="1" pmos_size="2"/>
<device_technology device_model_name="logic"/>
<input_buffer exist="false"/>
<output_buffer exist="false"/>
<port type="input" prefix="in" size="1"/>
<port type="input" prefix="sel" size="1"/>
<port type="input" prefix="selb" size="1"/>
<port type="output" prefix="out" size="1"/>
<delay_matrix type="rise" in_port="in sel selb" out_port="out">
10e-12 5e-12 5e-12
</delay_matrix>
<delay_matrix type="fall" in_port="in sel selb" out_port="out">
10e-12 5e-12 5e-12
</delay_matrix>
</circuit_model>
<circuit_model type="chan_wire" name="chan_segment" prefix="track_seg" is_default="true">
<design_technology type="cmos"/>
<input_buffer exist="false"/>
<output_buffer exist="false"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<wire_param model_type="pi" R="101" C="22.5e-15" num_level="1"/> <!-- model_type could be T, res_val and cap_val DON'T CARE -->
</circuit_model>
<circuit_model type="wire" name="direct_interc" prefix="direct_interc" is_default="true">
<design_technology type="cmos"/>
<input_buffer exist="false"/>
<output_buffer exist="false"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<wire_param model_type="pi" R="0" C="0" num_level="1"/> <!-- model_type could be T, res_val cap_val should be defined -->
</circuit_model>
<circuit_model type="mux" name="mux_2level" prefix="mux_2level" dump_structural_verilog="true">
<design_technology type="cmos" structure="multi_level" num_level="2" add_const_input="true" const_input_val="1"/>
<input_buffer exist="true" circuit_model_name="INVTX1"/>
<output_buffer exist="true" circuit_model_name="INVTX1"/>
<pass_gate_logic circuit_model_name="TGATE"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<port type="sram" prefix="sram" size="1"/>
</circuit_model>
<circuit_model type="mux" name="mux_2level_tapbuf" prefix="mux_2level_tapbuf" dump_structural_verilog="true">
<design_technology type="cmos" structure="multi_level" num_level="2" add_const_input="true" const_input_val="1"/>
<input_buffer exist="true" circuit_model_name="INVTX1"/>
<output_buffer exist="true" circuit_model_name="tap_buf4"/>
<pass_gate_logic circuit_model_name="TGATE"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<port type="sram" prefix="sram" size="1"/>
</circuit_model>
<circuit_model type="mux" name="mux_1level_tapbuf" prefix="mux_1level_tapbuf" is_default="true" dump_structural_verilog="true">
<design_technology type="cmos" structure="one_level" add_const_input="true" const_input_val="1"/>
<input_buffer exist="true" circuit_model_name="INVTX1"/>
<output_buffer exist="true" circuit_model_name="tap_buf4"/>
<pass_gate_logic circuit_model_name="TGATE"/>
<port type="input" prefix="in" size="1"/>
<port type="output" prefix="out" size="1"/>
<port type="sram" prefix="sram" size="1"/>
</circuit_model>
<!--DFF subckt ports should be defined as <D> <Q> <CLK> <RESET> <SET> -->
<circuit_model type="ff" name="DFFSRQ" prefix="DFFSRQ" spice_netlist="${OPENFPGA_PATH}/openfpga_flow/openfpga_cell_library/spice/dff.sp" verilog_netlist="${OPENFPGA_PATH}/openfpga_flow/openfpga_cell_library/verilog/dff.v">
<design_technology type="cmos"/>
<input_buffer exist="true" circuit_model_name="INVTX1"/>
<output_buffer exist="true" circuit_model_name="INVTX1"/>
<port type="input" prefix="D" lib_name="D" size="1"/>
<port type="input" prefix="set" lib_name="SET" size="1" is_global="true" default_val="0" is_set="true"/>
<port type="input" prefix="reset" lib_name="RST" size="1" is_global="true" default_val="0" is_reset="true"/>
<port type="output" prefix="Q" lib_name="Q" size="1"/>
<port type="clock" prefix="clk" lib_name="CK" size="1" is_global="true" default_val="0" />
</circuit_model>
<circuit_model type="lut" name="lut4" prefix="lut4" dump_structural_verilog="true">
<design_technology type="cmos"/>
<input_buffer exist="true" circuit_model_name="INVTX1"/>
<output_buffer exist="true" circuit_model_name="INVTX1"/>
<lut_input_inverter exist="true" circuit_model_name="INVTX1"/>
<lut_input_buffer exist="true" circuit_model_name="buf4"/>
<pass_gate_logic circuit_model_name="TGATE"/>
<port type="input" prefix="in" size="4"/>
<port type="output" prefix="out" size="1"/>
<port type="sram" prefix="sram" size="16"/>
</circuit_model>
<!--Scan-chain DFF subckt ports should be defined as <D> <Q> <Qb> <CLK> <RESET> <SET> -->
<circuit_model type="sram" name="SRAM" prefix="SRAM" spice_netlist="${OPENFPGA_PATH}/openfpga_flow/openfpga_cell_library/spice/sram.sp" verilog_netlist="${OPENFPGA_PATH}/openfpga_flow/openfpga_cell_library/verilog/sram.v">
<design_technology type="cmos"/>
<input_buffer exist="true" circuit_model_name="INVTX1"/>
<output_buffer exist="true" circuit_model_name="INVTX1"/>
<port type="bl" prefix="bl" lib_name="D" size="1"/>
<port type="wl" prefix="wl" lib_name="WE" size="1"/>
<port type="output" prefix="out" lib_name="Q" size="1"/>
<port type="output" prefix="outb" lib_name="QN" size="1"/>
</circuit_model>
<circuit_model type="iopad" name="GPIO" prefix="GPIO" spice_netlist="${OPENFPGA_PATH}/openfpga_flow/openfpga_cell_library/spice/gpio.sp" verilog_netlist="${OPENFPGA_PATH}/openfpga_flow/openfpga_cell_library/verilog/gpio.v">
<design_technology type="cmos"/>
<input_buffer exist="true" circuit_model_name="INVTX1"/>
<output_buffer exist="true" circuit_model_name="INVTX1"/>
<port type="inout" prefix="PAD" size="1" is_global="true" is_io="true" is_data_io="true"/>
<port type="sram" prefix="DIR" size="1" mode_select="true" circuit_model_name="SRAM" default_val="1"/>
<port type="input" prefix="outpad" lib_name="A" size="1"/>
<port type="output" prefix="inpad" lib_name="Y" size="1"/>
</circuit_model>
</circuit_library>
<configuration_protocol>
<organization type="ql_memory_bank" circuit_model_name="SRAM"/>
</configuration_protocol>
<connection_block>
<switch name="ipin_cblock" circuit_model_name="mux_2level_tapbuf"/>
</connection_block>
<switch_block>
<switch name="0" circuit_model_name="mux_2level_tapbuf"/>
</switch_block>
<routing_segment>
<segment name="L4" circuit_model_name="chan_segment"/>
</routing_segment>
<pb_type_annotations>
<!-- physical pb_type binding in complex block IO -->
<pb_type name="io" physical_mode_name="physical" idle_mode_name="inpad"/>
<pb_type name="io[physical].iopad" circuit_model_name="GPIO" mode_bits="1"/>
<pb_type name="io[inpad].inpad" physical_pb_type_name="io[physical].iopad" mode_bits="1"/>
<pb_type name="io[outpad].outpad" physical_pb_type_name="io[physical].iopad" mode_bits="0"/>
<!-- End physical pb_type binding in complex block IO -->
<!-- physical pb_type binding in complex block CLB -->
<!-- physical mode will be the default mode if not specified -->
<pb_type name="clb">
<!-- Binding interconnect to circuit models as their physical implementation, if not defined, we use the default model -->
<interconnect name="crossbar" circuit_model_name="mux_2level"/>
</pb_type>
<pb_type name="clb.fle[n1_lut4].ble4.lut4" circuit_model_name="lut4"/>
<pb_type name="clb.fle[n1_lut4].ble4.ff" circuit_model_name="DFFSRQ"/>
<!-- End physical pb_type binding in complex block IO -->
</pb_type_annotations>
</openfpga_architecture>

3
openfpga_flow/regression_test_scripts/basic_reg_test.sh
View File

@ -53,6 +53,9 @@ run-task basic_tests/full_testbench/smart_fast_memory_bank --debug --show_thread
run-task basic_tests/full_testbench/smart_fast_multi_region_memory_bank --debug --show_thread_logs run-task basic_tests/full_testbench/smart_fast_multi_region_memory_bank --debug --show_thread_logs
run-task basic_tests/preconfig_testbench/memory_bank --debug --show_thread_logs run-task basic_tests/preconfig_testbench/memory_bank --debug --show_thread_logs
echo -e "Testing physical design friendly memory bank configuration protocol of a K4N4 FPGA";
run-task basic_tests/full_testbench/ql_memory_bank --debug --show_thread_logs
echo -e "Testing testbenches without self checking features"; echo -e "Testing testbenches without self checking features";
run-task basic_tests/full_testbench/full_testbench_without_self_checking --debug --show_thread_logs run-task basic_tests/full_testbench/full_testbench_without_self_checking --debug --show_thread_logs
run-task basic_tests/preconfig_testbench/preconfigured_testbench_without_self_checking --debug --show_thread_logs run-task basic_tests/preconfig_testbench/preconfigured_testbench_without_self_checking --debug --show_thread_logs

44
openfpga_flow/tasks/basic_tests/full_testbench/ql_memory_bank/config/task.conf Normal file
View File

@ -0,0 +1,44 @@
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
# Configuration file for running experiments
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
# timeout_each_job : FPGA Task script splits fpga flow into multiple jobs
# Each job execute fpga_flow script on combination of architecture & benchmark
# timeout_each_job is timeout for each job
# = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = = =
[GENERAL]
run_engine=openfpga_shell
power_tech_file = ${PATH:OPENFPGA_PATH}/openfpga_flow/tech/PTM_45nm/45nm.xml
power_analysis = true
spice_output=false
verilog_output=true
timeout_each_job = 20*60
fpga_flow=yosys_vpr
[OpenFPGA_SHELL]
openfpga_shell_template=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_shell_scripts/write_full_testbench_example_script.openfpga
openfpga_arch_file=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_arch/k4_N4_40nm_qlbank_openfpga.xml
openfpga_sim_setting_file=${PATH:OPENFPGA_PATH}/openfpga_flow/openfpga_simulation_settings/auto_sim_openfpga.xml
openfpga_vpr_device_layout=
openfpga_fast_configuration=
[ARCHITECTURES]
arch0=${PATH:OPENFPGA_PATH}/openfpga_flow/vpr_arch/k4_N4_tileable_40nm.xml
[BENCHMARKS]
bench0=${PATH:OPENFPGA_PATH}/openfpga_flow/benchmarks/micro_benchmark/and2/and2.v
bench1=${PATH:OPENFPGA_PATH}/openfpga_flow/benchmarks/micro_benchmark/or2/or2.v
bench2=${PATH:OPENFPGA_PATH}/openfpga_flow/benchmarks/micro_benchmark/and2_latch/and2_latch.v
[SYNTHESIS_PARAM]
bench0_top = and2
bench0_chan_width = 300
bench1_top = or2
bench1_chan_width = 300
bench2_top = and2_latch
bench2_chan_width = 300
[SCRIPT_PARAM_MIN_ROUTE_CHAN_WIDTH]
end_flow_with_test=