riscv
/
yosys
mirror of https://github.com/YosysHQ/yosys.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
yosys/techlibs/intel_alm/synth_intel_alm.cc

301 lines
10 KiB
C++
Raw Normal View History

synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
/*
* yosys -- Yosys Open SYnthesis Suite
*
* Copyright (C) 2012 Claire Wolf <claire@symbioticeda.com>
* Copyright (C) 2019 Dan Ravensloft <dan.ravensloft@gmail.com>
*
* Permission to use, copy, modify, and/or distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*
*/
#include "kernel/celltypes.h"
#include "kernel/log.h"
#include "kernel/register.h"
#include "kernel/rtlil.h"
USING_YOSYS_NAMESPACE
PRIVATE_NAMESPACE_BEGIN
struct SynthIntelALMPass : public ScriptPass {
SynthIntelALMPass() : ScriptPass("synth_intel_alm", "synthesis for ALM-based Intel (Altera) FPGAs.") {}
Use C++11 final/override keywords.
2020-06-18 18:34:52 -05:00
void help() override
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
{
// |---v---|---v---|---v---|---v---|---v---|---v---|---v---|---v---|---v---|---v---|
log("\n");
log(" synth_intel_alm [options]\n");
log("\n");
log("This command runs synthesis for ALM-based Intel FPGAs.\n");
log("\n");
log(" -top <module>\n");
intel_alm: ABC9 sequential optimisations
2020-05-23 06:52:13 -05:00
log(" use the specified module as top module\n");
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
log("\n");
log(" -family <family>\n");
log(" target one of:\n");
log(" \"cyclonev\" - Cyclone V (default)\n");
log(" \"cyclone10gx\" - Cyclone 10GX\n");
log("\n");
log(" -vqm <file>\n");
log(" write the design to the specified Verilog Quartus Mapping File. Writing of an\n");
log(" output file is omitted if this parameter is not specified. Implies -quartus.\n");
log("\n");
intel_alm: ABC9 sequential optimisations
2020-05-23 06:52:13 -05:00
log(" -noflatten\n");
log(" do not flatten design before synthesis; useful for per-module area statistics\n");
log("\n");
log(" -quartus\n");
log(" output a netlist using Quartus cells instead of MISTRAL_* cells\n");
log("\n");
log(" -dff\n");
log(" pass DFFs to ABC to perform sequential logic optimisations (EXPERIMENTAL)\n");
log("\n");
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
log(" -run <from_label>:<to_label>\n");
log(" only run the commands between the labels (see below). an empty\n");
log(" from label is synonymous to 'begin', and empty to label is\n");
log(" synonymous to the end of the command list.\n");
log("\n");
log(" -nolutram\n");
log(" do not use LUT RAM cells in output netlist\n");
log("\n");
log(" -nobram\n");
log(" do not use block RAM cells in output netlist\n");
log("\n");
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
log(" -nodsp\n");
log(" do not map multipliers to MISTRAL_MUL cells\n");
log("\n");
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
log("The following commands are executed by this synthesis command:\n");
help_script();
log("\n");
}
string top_opt, family_opt, bram_type, vout_file;
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
bool flatten, quartus, nolutram, nobram, dff, nodsp;
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
Use C++11 final/override keywords.
2020-06-18 18:34:52 -05:00
void clear_flags() override
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
{
top_opt = "-auto-top";
family_opt = "cyclonev";
bram_type = "m10k";
vout_file = "";
flatten = true;
quartus = false;
nolutram = false;
nobram = false;
intel_alm: ABC9 sequential optimisations
2020-05-23 06:52:13 -05:00
dff = false;
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
nodsp = false;
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
}
Use C++11 final/override keywords.
2020-06-18 18:34:52 -05:00
void execute(std::vector<std::string> args, RTLIL::Design *design) override
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
{
string run_from, run_to;
clear_flags();
size_t argidx;
for (argidx = 1; argidx < args.size(); argidx++) {
if (args[argidx] == "-family" && argidx + 1 < args.size()) {
family_opt = args[++argidx];
continue;
}
if (args[argidx] == "-top" && argidx + 1 < args.size()) {
top_opt = "-top " + args[++argidx];
continue;
}
if (args[argidx] == "-vqm" && argidx + 1 < args.size()) {
quartus = true;
vout_file = args[++argidx];
continue;
}
if (args[argidx] == "-run" && argidx + 1 < args.size()) {
size_t pos = args[argidx + 1].find(':');
if (pos == std::string::npos)
break;
run_from = args[++argidx].substr(0, pos);
run_to = args[argidx].substr(pos + 1);
continue;
}
if (args[argidx] == "-quartus") {
quartus = true;
continue;
}
if (args[argidx] == "-nolutram") {
nolutram = true;
continue;
}
if (args[argidx] == "-nobram") {
nobram = true;
continue;
}
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
if (args[argidx] == "-nodsp") {
nodsp = true;
continue;
}
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
if (args[argidx] == "-noflatten") {
flatten = false;
continue;
}
intel_alm: ABC9 sequential optimisations
2020-05-23 06:52:13 -05:00
if (args[argidx] == "-dff") {
dff = true;
continue;
}
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
break;
}
extra_args(args, argidx, design);
if (!design->full_selection())
log_cmd_error("This command only operates on fully selected designs!\n");
if (family_opt == "cyclonev") {
bram_type = "m10k";
} else if (family_opt == "cyclone10gx") {
bram_type = "m20k";
} else {
log_cmd_error("Invalid family specified: '%s'\n", family_opt.c_str());
}
log_header(design, "Executing SYNTH_INTEL_ALM pass.\n");
log_push();
run_script(design, run_from, run_to);
log_pop();
}
Use C++11 final/override keywords.
2020-06-18 18:34:52 -05:00
void script() override
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
{
if (help_mode) {
family_opt = "<family>";
bram_type = "<bram_type>";
}
if (check_label("begin")) {
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
if (family_opt == "cyclonev")
run(stringf("read_verilog -sv -lib +/intel/%s/cells_sim.v", family_opt.c_str()));
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
run(stringf("read_verilog -specify -lib -D %s +/intel_alm/common/alm_sim.v", family_opt.c_str()));
run(stringf("read_verilog -specify -lib -D %s +/intel_alm/common/dff_sim.v", family_opt.c_str()));
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
run(stringf("read_verilog -specify -lib -D %s +/intel_alm/common/dsp_sim.v", family_opt.c_str()));
intel_alm: direct LUTRAM cell instantiation By instantiating the LUTRAM cell directly, we avoid a trip through altsyncram, which speeds up Quartus synthesis time. This also gives a little more flexibility, as Yosys can build RAMs out of individual 32x1 LUTRAM cells. While working on this, I discovered that the mem_init0 parameter of <family>_mlab_cell gets ignored by Quartus.
2020-04-16 06:24:04 -05:00
run(stringf("read_verilog -specify -lib -D %s +/intel_alm/common/mem_sim.v", family_opt.c_str()));
intel_alm: add $__ prefix to MISTRAL_FF_SYNCONLY
2020-05-25 17:15:20 -05:00
run(stringf("read_verilog -specify -lib -D %s -icells +/intel_alm/common/abc9_model.v", family_opt.c_str()));
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
// Misc and common cells
run("read_verilog -lib +/intel/common/altpll_bb.v");
run("read_verilog -lib +/intel_alm/common/megafunction_bb.v");
run(stringf("hierarchy -check %s", help_mode ? "-top <top>" : top_opt.c_str()));
}
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
if (check_label("coarse")) {
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
run("proc");
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
if (flatten || help_mode)
run("flatten", "(skip if -noflatten)");
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
run("tribuf -logic");
run("deminout");
intel_alm: DSP inference
2020-04-25 11:25:59 -05:00
run("opt_expr");
run("opt_clean");
run("check");
run("opt");
run("wreduce");
run("peepopt");
run("opt_clean");
run("share");
run("techmap -map +/cmp2lut.v -D LUT_WIDTH=6");
run("opt_expr");
run("opt_clean");
if (help_mode) {
run("techmap -map +/mul2dsp.v [...]", "(unless -nodsp)");
} else if (!nodsp) {
// Cyclone V supports 9x9 multiplication, Cyclone 10 GX does not.
run("techmap -map +/mul2dsp.v -D DSP_A_MAXWIDTH=27 -D DSP_B_MAXWIDTH=27 -D DSP_A_MINWIDTH=19 -D DSP_B_MINWIDTH=19 -D DSP_SIGNEDONLY -D DSP_NAME=__MUL27X27");
run("chtype -set $mul t:$__soft_mul");
if (family_opt == "cyclonev") {
run("techmap -map +/mul2dsp.v -D DSP_A_MAXWIDTH=18 -D DSP_B_MAXWIDTH=18 -D DSP_A_MINWIDTH=10 -D DSP_B_MINWIDTH=10 -D DSP_SIGNEDONLY -D DSP_NAME=__MUL18X18");
run("chtype -set $mul t:$__soft_mul");
run("techmap -map +/mul2dsp.v -D DSP_A_MAXWIDTH=9 -D DSP_B_MAXWIDTH=9 -D DSP_A_MINWIDTH=4 -D DSP_B_MINWIDTH=4 -D DSP_SIGNEDONLY -D DSP_NAME=__MUL9X9");
run("chtype -set $mul t:$__soft_mul");
} else if (family_opt == "cyclone10gx") {
run("techmap -map +/mul2dsp.v -D DSP_A_MAXWIDTH=18 -D DSP_B_MAXWIDTH=18 -D DSP_A_MINWIDTH=4 -D DSP_B_MINWIDTH=4 -D DSP_SIGNEDONLY -D DSP_NAME=__MUL18X18");
run("chtype -set $mul t:$__soft_mul");
}
}
run("alumacc");
run("techmap -map +/intel_alm/common/arith_alm_map.v -map +/intel_alm/common/dsp_map.v");
run("opt");
run("fsm");
run("opt -fast");
run("memory -nomap");
run("opt_clean");
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
}
if (!nobram && check_label("map_bram", "(skip if -nobram)")) {
run(stringf("memory_bram -rules +/intel_alm/common/bram_%s.txt", bram_type.c_str()));
intel_alm: direct M10K instantiation
2020-06-11 16:25:04 -05:00
if (help_mode || bram_type != "m10k")
run(stringf("techmap -map +/intel_alm/common/bram_%s_map.v", bram_type.c_str()));
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
}
if (!nolutram && check_label("map_lutram", "(skip if -nolutram)")) {
run("memory_bram -rules +/intel_alm/common/lutram_mlab.txt", "(for Cyclone V / Cyclone 10GX)");
}
if (check_label("map_ffram")) {
run("memory_map");
run("opt -full");
}
if (check_label("map_ffs")) {
synth_intel_alm: Use dfflegalize.
2020-06-30 20:31:34 -05:00
run("techmap");
intel_alm: fix DFFE matching
2020-06-11 12:06:39 -05:00
run("dff2dffe");
synth_intel_alm: Use dfflegalize.
2020-06-30 20:31:34 -05:00
run("dfflegalize -cell $_DFFE_PN0P_ 0 -cell $_SDFFCE_PP0P_ 0");
run("techmap -map +/intel_alm/common/dff_map.v");
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
run("opt -full -undriven -mux_undef");
run("clean -purge");
}
if (check_label("map_luts")) {
intel_alm: ABC9 sequential optimisations
2020-05-23 06:52:13 -05:00
run("techmap -map +/intel_alm/common/abc9_map.v");
run(stringf("abc9 %s -maxlut 6 -W 200", help_mode ? "[-dff]" : dff ? "-dff" : ""));
run("techmap -map +/intel_alm/common/abc9_unmap.v");
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
run("techmap -map +/intel_alm/common/alm_map.v");
run("opt -fast");
run("autoname");
run("clean");
}
if (check_label("check")) {
run("hierarchy -check");
run("stat");
run("check");
}
if (check_label("quartus")) {
if (quartus || help_mode) {
intel_alm: work around a Quartus ICE
2020-04-22 18:56:49 -05:00
// Quartus ICEs if you have a wire which has `[]` in its name,
// which Yosys produces when building memories out of flops.
run("rename -hide w:*[* w:*]*");
// VQM mode does not support 'x, so replace those with zero.
synth_intel_alm: VQM support
2020-04-15 08:28:35 -05:00
run("setundef -zero");
intel_alm: work around a Quartus ICE
2020-04-22 18:56:49 -05:00
// VQM mode does not support multi-bit constant assignments
// (e.g. 2'b00 is an error), so as a workaround use references
// to constant driver cells, which Quartus accepts.
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
run("hilomap -singleton -hicell __MISTRAL_VCC Q -locell __MISTRAL_GND Q");
intel_alm: work around a Quartus ICE
2020-04-22 18:56:49 -05:00
// Rename from Yosys-internal MISTRAL_* cells to Quartus cells.
intel_alm: cleanup duplication
2020-04-23 16:44:29 -05:00
run(stringf("techmap -D %s -map +/intel_alm/common/quartus_rename.v", family_opt.c_str()));
synth_intel_alm: alternative synthesis for Intel FPGAs By operating at a layer of abstraction over the rather clumsy Intel primitives, we can avoid special hacks like `dffinit -highlow` in favour of simple techmapping. This also makes the primitives much easier to manipulate, and more descriptive (no more cyclonev_lcell_comb to mean anything from a LUT2 to a LUT6).
2019-11-19 04:19:00 -06:00
}
}
if (check_label("vqm")) {
if (!vout_file.empty() || help_mode) {
run(stringf("write_verilog -attr2comment -defparam -nohex -decimal %s", help_mode ? "<file-name>" : vout_file.c_str()));
}
}
}
} SynthIntelALMPass;
PRIVATE_NAMESPACE_END