stefan
/
mlhw
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
Browse Source

synthesized fma packages

main
stefan 3 hours ago
commit
dc40d84683
11 changed files with 781 additions and 0 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 1
      .gitignore
  2. 1
      .python-version
  3. 6
      README.md
  4. 330
      flake.lock
  5. 212
      flake.nix
  6. 11
      fma_fixed.v
  7. 69
      fma_float16.sv
  8. 69
      fma_float32.sv
  9. 69
      fma_float64.sv
  10. 6
      main.py
  11. 7
      pyproject.toml

1
.gitignore vendored
Unescape View File

@ -0,0 +1 @@
result

1
.python-version
Unescape View File

@ -0,0 +1 @@
3.14

6
README.md
Unescape View File

@ -0,0 +1,6 @@
Machine Learning Hardware Modules
=================================
Collection of various hardware components used for machine learning accelerator research.

330
flake.lock
Unescape View File

@ -0,0 +1,330 @@
{
"nodes": {
"benchmark-circuits": {
"inputs": {
"librelane": "librelane",
"nixpkgs": "nixpkgs_2"
},
"locked": {
"lastModified": 1779683213,
"narHash": "sha256-4VTmok0G04LHmz/gOw1D6/NJHW4gq85YGKEuPr1RgeM=",
"owner": "s-holst",
"repo": "benchmark-circuits",
"rev": "72f22e850f97252a8ae84b0f76a6f9472f74de38",
"type": "github"
},
"original": {
"owner": "s-holst",
"repo": "benchmark-circuits",
"type": "github"
}
},
"ciel": {
"inputs": {
"nix-eda": [
"benchmark-circuits",
"librelane",
"nix-eda"
]
},
"locked": {
"lastModified": 1764091696,
"narHash": "sha256-AWbkHL0zO3tD0mE3dZIcj8mVND7o3imTxOpEfOtlRDI=",
"owner": "fossi-foundation",
"repo": "ciel",
"rev": "afcb23d368614ffa1e7e96584ed33f839c71c576",
"type": "github"
},
"original": {
"owner": "fossi-foundation",
"repo": "ciel",
"type": "github"
}
},
"ciel_2": {
"inputs": {
"nix-eda": [
"librelane",
"nix-eda"
]
},
"locked": {
"lastModified": 1764091696,
"narHash": "sha256-AWbkHL0zO3tD0mE3dZIcj8mVND7o3imTxOpEfOtlRDI=",
"owner": "fossi-foundation",
"repo": "ciel",
"rev": "afcb23d368614ffa1e7e96584ed33f839c71c576",
"type": "github"
},
"original": {
"owner": "fossi-foundation",
"repo": "ciel",
"type": "github"
}
},
"cvfpu-src": {
"flake": false,
"locked": {
"lastModified": 1778159595,
"narHash": "sha256-GV8i3+GgsshXVTRrivl9EmTWupzckVWE4DiYBuyD1fs=",
"ref": "refs/heads/develop",
"rev": "106251e502747f17d931a20db0bdbab9e1a6c2ff",
"revCount": 304,
"submodules": true,
"type": "git",
"url": "https://github.com/openhwgroup/cvfpu.git"
},
"original": {
"submodules": true,
"type": "git",
"url": "https://github.com/openhwgroup/cvfpu.git"
}
},
"devshell": {
"inputs": {
"nixpkgs": [
"benchmark-circuits",
"librelane",
"nix-eda",
"nixpkgs"
]
},
"locked": {
"lastModified": 1768818222,
"narHash": "sha256-460jc0+CZfyaO8+w8JNtlClB2n4ui1RbHfPTLkpwhU8=",
"owner": "numtide",
"repo": "devshell",
"rev": "255a2b1725a20d060f566e4755dbf571bbbb5f76",
"type": "github"
},
"original": {
"owner": "numtide",
"repo": "devshell",
"type": "github"
}
},
"devshell_2": {
"inputs": {
"nixpkgs": [
"librelane",
"nix-eda",
"nixpkgs"
]
},
"locked": {
"lastModified": 1768818222,
"narHash": "sha256-460jc0+CZfyaO8+w8JNtlClB2n4ui1RbHfPTLkpwhU8=",
"owner": "numtide",
"repo": "devshell",
"rev": "255a2b1725a20d060f566e4755dbf571bbbb5f76",
"type": "github"
},
"original": {
"owner": "numtide",
"repo": "devshell",
"type": "github"
}
},
"flake-compat": {
"locked": {
"lastModified": 1733328505,
"narHash": "sha256-NeCCThCEP3eCl2l/+27kNNK7QrwZB1IJCrXfrbv5oqU=",
"rev": "ff81ac966bb2cae68946d5ed5fc4994f96d0ffec",
"revCount": 69,
"type": "tarball",
"url": "https://api.flakehub.com/f/pinned/edolstra/flake-compat/1.1.0/01948eb7-9cba-704f-bbf3-3fa956735b52/source.tar.gz"
},
"original": {
"type": "tarball",
"url": "https://flakehub.com/f/edolstra/flake-compat/1.tar.gz"
}
},
"flake-compat_2": {
"locked": {
"lastModified": 1733328505,
"narHash": "sha256-NeCCThCEP3eCl2l/+27kNNK7QrwZB1IJCrXfrbv5oqU=",
"rev": "ff81ac966bb2cae68946d5ed5fc4994f96d0ffec",
"revCount": 69,
"type": "tarball",
"url": "https://api.flakehub.com/f/pinned/edolstra/flake-compat/1.1.0/01948eb7-9cba-704f-bbf3-3fa956735b52/source.tar.gz"
},
"original": {
"type": "tarball",
"url": "https://flakehub.com/f/edolstra/flake-compat/1.tar.gz"
}
},
"librelane": {
"inputs": {
"ciel": "ciel",
"devshell": "devshell",
"flake-compat": "flake-compat",
"nix-eda": "nix-eda"
},
"locked": {
"lastModified": 1775150546,
"narHash": "sha256-JxWuaOBhkjjw4sp7l++QF+0EzGIhPAOaJcKwwmdTg+w=",
"owner": "librelane",
"repo": "librelane",
"rev": "c82049f0234a49bbf72ebcc89539529f31d8997e",
"type": "github"
},
"original": {
"owner": "librelane",
"ref": "3.0.2",
"repo": "librelane",
"type": "github"
}
},
"librelane_2": {
"inputs": {
"ciel": "ciel_2",
"devshell": "devshell_2",
"flake-compat": "flake-compat_2",
"nix-eda": "nix-eda_2"
},
"locked": {
"lastModified": 1775150546,
"narHash": "sha256-JxWuaOBhkjjw4sp7l++QF+0EzGIhPAOaJcKwwmdTg+w=",
"owner": "librelane",
"repo": "librelane",
"rev": "c82049f0234a49bbf72ebcc89539529f31d8997e",
"type": "github"
},
"original": {
"owner": "librelane",
"ref": "3.0.2",
"repo": "librelane",
"type": "github"
}
},
"nix-eda": {
"inputs": {
"nixpkgs": "nixpkgs"
},
"locked": {
"lastModified": 1773918136,
"narHash": "sha256-nSKBMGP8/ZC7qB3Lzd+FwM8REqOxlh8wpYDf2hlK6Gg=",
"owner": "fossi-foundation",
"repo": "nix-eda",
"rev": "8f990fb77529c09e540e453cd836af9930ec58db",
"type": "github"
},
"original": {
"owner": "fossi-foundation",
"ref": "6.11.0",
"repo": "nix-eda",
"type": "github"
}
},
"nix-eda_2": {
"inputs": {
"nixpkgs": "nixpkgs_3"
},
"locked": {
"lastModified": 1773918136,
"narHash": "sha256-nSKBMGP8/ZC7qB3Lzd+FwM8REqOxlh8wpYDf2hlK6Gg=",
"owner": "fossi-foundation",
"repo": "nix-eda",
"rev": "8f990fb77529c09e540e453cd836af9930ec58db",
"type": "github"
},
"original": {
"owner": "fossi-foundation",
"ref": "6.11.0",
"repo": "nix-eda",
"type": "github"
}
},
"nixpkgs": {
"locked": {
"lastModified": 1766201043,
"narHash": "sha256-eplAP+rorKKd0gNjV3rA6+0WMzb1X1i16F5m5pASnjA=",
"owner": "nixos",
"repo": "nixpkgs",
"rev": "b3aad468604d3e488d627c0b43984eb60e75e782",
"type": "github"
},
"original": {
"owner": "nixos",
"ref": "nixos-25.11",
"repo": "nixpkgs",
"type": "github"
}
},
"nixpkgs_2": {
"locked": {
"lastModified": 1777268161,
"narHash": "sha256-bxrdOn8SCOv8tN4JbTF/TXq7kjo9ag4M+C8yzzIRYbE=",
"owner": "NixOS",
"repo": "nixpkgs",
"rev": "1c3fe55ad329cbcb28471bb30f05c9827f724c76",
"type": "github"
},
"original": {
"owner": "NixOS",
"ref": "nixos-unstable",
"repo": "nixpkgs",
"type": "github"
}
},
"nixpkgs_3": {
"locked": {
"lastModified": 1766201043,
"narHash": "sha256-eplAP+rorKKd0gNjV3rA6+0WMzb1X1i16F5m5pASnjA=",
"owner": "nixos",
"repo": "nixpkgs",
"rev": "b3aad468604d3e488d627c0b43984eb60e75e782",
"type": "github"
},
"original": {
"owner": "nixos",
"ref": "nixos-25.11",
"repo": "nixpkgs",
"type": "github"
}
},
"nixpkgs_4": {
"locked": {
"lastModified": 1779508470,
"narHash": "sha256-Ap9KJX+5xHIn3bPIpfNgT6MEXdAECECwo4/rmlQD74M=",
"owner": "NixOS",
"repo": "nixpkgs",
"rev": "29916453413845e54a65b8a1cf996842300cd299",
"type": "github"
},
"original": {
"owner": "NixOS",
"ref": "nixos-unstable",
"repo": "nixpkgs",
"type": "github"
}
},
"root": {
"inputs": {
"benchmark-circuits": "benchmark-circuits",
"cvfpu-src": "cvfpu-src",
"librelane": "librelane_2",
"nixpkgs": "nixpkgs_4",
"sv2v-src": "sv2v-src"
}
},
"sv2v-src": {
"flake": false,
"locked": {
"lastModified": 1774651589,
"narHash": "sha256-ziwLw1/S4wbnqml/AnN/yerOJJ3VOfRc3dZa8cmEaD0=",
"owner": "zachjs",
"repo": "sv2v",
"rev": "6662fa5da71f87797598060f17728b284b99a9fc",
"type": "github"
},
"original": {
"owner": "zachjs",
"repo": "sv2v",
"type": "github"
}
}
},
"root": "root",
"version": 7
}

212
flake.nix
Unescape View File

@ -0,0 +1,212 @@
{
description = "Packages for synthesizing machine learning hardware modules";
inputs = {
nixpkgs.url = "github:NixOS/nixpkgs/nixos-unstable";
librelane.url = "github:librelane/librelane/3.0.2";
benchmark-circuits.url = "github:s-holst/benchmark-circuits";
sv2v-src = {
url = "github:zachjs/sv2v";
flake = false;
};
cvfpu-src = {
url = "git+https://github.com/openhwgroup/cvfpu.git?submodules=1";
flake = false;
};
};
outputs = { self, nixpkgs, librelane, benchmark-circuits, sv2v-src, cvfpu-src }:
let
systems = [ "x86_64-linux" "aarch64-linux" "x86_64-darwin" "aarch64-darwin" ];
forAllSystems = nixpkgs.lib.genAttrs systems;
in {
packages = forAllSystems (system:
let
pkgs = nixpkgs.legacyPackages.${system};
librelane-pkg = librelane.packages.${system}.default;
gsclib = benchmark-circuits.packages.${system}.iwls-gsclib;
sky130-pdk = benchmark-circuits.packages.${system}.sky130-pdk;
sv2v = pkgs.haskell.lib.compose.justStaticExecutables (pkgs.haskell.packages.ghc910.callCabal2nix "sv2v" sv2v-src { });
run-sv2v = {data_width}:
''
sed -e 's/default disable iff/\/\/default disable iff/' <$src/src/common_cells/src/rr_arb_tree.sv >rr_arb_tree_patched.sv
sv2v \
-I$src/src/common_cells/include \
$src/src/common_cells/src/cf_math_pkg.sv \
$src/src/common_cells/src/lzc.sv \
rr_arb_tree_patched.sv \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/clk/rtl/gated_clk_cell.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_ctrl.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_ff1.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_pack_single.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_prepare.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_round_single.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_special.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_srt_single.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fdsu/rtl/pa_fdsu_top.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fpu/rtl/pa_fpu_dp.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fpu/rtl/pa_fpu_frbus.v \
$src/vendor/opene906/E906_RTL_FACTORY/gen_rtl/fpu/rtl/pa_fpu_src_type.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_ctrl.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_double.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_ff1.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_pack.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_prepare.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_round.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_scalar_dp.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_srt_radix16_bound_table.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_srt_radix16_with_sqrt.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_srt.v \
$src/vendor/openc910/C910_RTL_FACTORY/gen_rtl/vfdsu/rtl/ct_vfdsu_top.v \
$src/src/fpnew_pkg.sv \
$src/src/fpnew_cast_multi.sv \
$src/src/fpnew_divsqrt_th_64_multi.sv \
$src/src/fpnew_opgroup_multifmt_slice.sv \
$src/src/fpnew_noncomp.sv \
$src/src/fpnew_rounding.sv \
$src/src/fpnew_classifier.sv \
$src/src/fpnew_fma.sv \
$srcExtra/fma_float${toString data_width}.sv | sed -e 's/\$finish/\/\/\$finish/; s/\(initial \)*\$display/\/\/\$display/' > top.v
'';
run-yosys = {verilog_file ? "top.v", data_width ? 0, sum_width ? 0}:
''
yosys <<EOF >out.log 2>&1
# Read design
read_verilog ${verilog_file}
hierarchy -check -top top ''
+ (if data_width > 0 then "-chparam DATA_WIDTH ${toString data_width}" else "")
+ (if sum_width > 0 then "-chparam SUM_WIDTH ${toString sum_width}" else "")
+ ''
# Coarse-grain synthesis
flatten; proc; memory; opt;
fsm; opt; # FSM pass re-codes states
wreduce; opt;
splitnets; opt;
# Technology mapping
techmap; opt; clean;
dfflibmap -liberty ${gsclib}/GSCLib_3.0.lib
abc -liberty ${gsclib}/GSCLib_3.0.lib
opt_clean
# Write netlist
write_verilog -noattr nl.v
EOF
'';
run-librelane = {verilog_file ? "top.v", data_width ? 0, sum_width ? 0, clock_period ? 20}:
''
mkdir top
cat > top/config.yaml << EOF
DESIGN_NAME: top
''
+ (if data_width > 0 then "SYNTH_PARAMETERS: [ 'DATA_WIDTH=${toString data_width}' "
+ (if sum_width > 0 then ", 'SUM_WIDTH=${toString sum_width}' ]\n" else "]\n") else "") +
''
VERILOG_FILES: [ ${verilog_file} ]
CLOCK_PERIOD: ${toString clock_period}
EOF
cat top/config.yaml
librelane --manual-pdk --pdk-root ${sky130-pdk} --run-tag run top/config.yaml
'';
mk_fma_fixed_gsclib = {data_width, sum_width}: pkgs.stdenv.mkDerivation {
name = "fma_fixed_${toString data_width}_${toString sum_width}_gsclib";
src = ./.;
unpackPhase = "true";
buildPhase = run-yosys { verilog_file = "$src/fma_fixed.v"; inherit data_width; inherit sum_width; };
installPhase =
''
ODIR=fma_fixed_${toString data_width}_${toString sum_width}_gsclib
mkdir -p $out/$ODIR/{log,nl}
cp out.log $out/$ODIR/log/top.log
cp nl.v $out/$ODIR/nl/top.nl.v
'';
buildInputs = [ pkgs.yosys ];
};
mk_fma_float_gsclib = {data_width}: pkgs.stdenv.mkDerivation {
name = "fma_float_${toString data_width}_gsclib";
src = cvfpu-src;
srcExtra = ./.;
unpackPhase = "true";
buildPhase = run-sv2v {inherit data_width;} + run-yosys {};
installPhase =
''
ODIR=fma_float_${toString data_width}_gsclib
mkdir -p $out/$ODIR/{log,nl}
cp out.log $out/$ODIR/log/top.log
cp nl.v $out/$ODIR/nl/top.nl.v
'';
buildInputs = [ pkgs.yosys sv2v ];
};
mk_fma_fixed_sky130 = {data_width, sum_width, clock_period} : pkgs.stdenv.mkDerivation {
name = "fma_fixed_${toString data_width}_${toString sum_width}_sky130";
src = ./.;
unpackPhase = "true";
buildPhase = run-librelane { verilog_file = "$src/fma_fixed.v"; inherit data_width; inherit sum_width; inherit clock_period;};
installPhase = ''
ODIR=fma_fixed_${toString data_width}_${toString sum_width}_sky130
mkdir -p $out/$ODIR
cp -r top/runs/run/final/. $out/$ODIR/
'';
buildInputs = [ librelane-pkg ];
};
mk_fma_float_sky130 = {data_width, clock_period}: pkgs.stdenv.mkDerivation {
name = "fma_float_${toString data_width}_sky130";
src = cvfpu-src;
srcExtra = ./.;
unpackPhase = "true";
buildPhase = run-sv2v {inherit data_width;} + run-librelane {inherit clock_period;};
installPhase =
''
ODIR=fma_float_${toString data_width}_sky130
mkdir -p $out/$ODIR
cp -r top/runs/run/final/. $out/$ODIR/
'';
buildInputs = [ librelane-pkg sv2v ];
};
in {
fma_fixed_8_24_gsclib = mk_fma_fixed_gsclib {data_width = 8; sum_width = 24; };
fma_fixed_16_40_gsclib = mk_fma_fixed_gsclib {data_width = 16; sum_width = 40; };
fma_fixed_32_72_gsclib = mk_fma_fixed_gsclib {data_width = 32; sum_width = 72; };
fma_fixed_8_24_sky130 = mk_fma_fixed_sky130 {data_width = 8; sum_width = 24; clock_period=20; };
fma_fixed_16_40_sky130 = mk_fma_fixed_sky130 {data_width = 16; sum_width = 40; clock_period=20; };
fma_fixed_32_72_sky130 = mk_fma_fixed_sky130 {data_width = 32; sum_width = 72; clock_period=20; };
fma_float_16_gsclib = mk_fma_float_gsclib {data_width = 16; };
fma_float_32_gsclib = mk_fma_float_gsclib {data_width = 32; };
fma_float_64_gsclib = mk_fma_float_gsclib {data_width = 64; };
fma_float_16_sky130 = mk_fma_float_sky130 {data_width = 16; clock_period=40; };
fma_float_32_sky130 = mk_fma_float_sky130 {data_width = 32; clock_period=50; };
fma_float_64_sky130 = mk_fma_float_sky130 {data_width = 64; clock_period=115; };
default = pkgs.symlinkJoin {
name = "fma_all";
paths = with self.packages.${system};
[
fma_fixed_8_24_gsclib
fma_fixed_16_40_gsclib
fma_fixed_32_72_gsclib
fma_fixed_8_24_sky130
fma_fixed_16_40_sky130
fma_fixed_32_72_sky130
fma_float_16_gsclib
fma_float_32_gsclib
fma_float_64_gsclib
fma_float_16_sky130
fma_float_32_sky130
fma_float_64_sky130
];
};
}
);
};
}

11
fma_fixed.v
Unescape View File

@ -0,0 +1,11 @@
module top #(
parameter DATA_WIDTH = 32,
parameter SUM_WIDTH = 72
) (
input wire signed [DATA_WIDTH-1:0] i_weight,
input wire signed [DATA_WIDTH-1:0] i_activation,
input wire signed [SUM_WIDTH-1:0] i_sum,
output wire signed [SUM_WIDTH-1:0] o_sum
);
assign o_sum = (i_weight * i_activation) + i_sum;
endmodule

69
fma_float16.sv
Unescape View File

@ -0,0 +1,69 @@
module top #(
parameter fpnew_pkg::fp_format_e FpFormat = fpnew_pkg::FP16,
parameter int unsigned NumPipeRegs = 0,
parameter fpnew_pkg::pipe_config_t PipeConfig = fpnew_pkg::BEFORE,
parameter type TagType = logic,
parameter type AuxType = logic,
localparam int unsigned WIDTH = 16,
localparam int unsigned ExtRegEnaWidth = 1
) (
input logic clk_i,
input logic rst_ni,
// Input signals
input logic [2:0][WIDTH-1:0] operands_i, // 3 operands
input logic [2:0] is_boxed_i, // 3 operands
input fpnew_pkg::roundmode_e rnd_mode_i,
input fpnew_pkg::operation_e op_i,
input logic op_mod_i,
input TagType tag_i,
input logic mask_i,
input AuxType aux_i,
// Input Handshake
input logic in_valid_i,
output logic in_ready_o,
input logic flush_i,
// Output signals
output logic [WIDTH-1:0] result_o,
output fpnew_pkg::status_t status_o,
output logic extension_bit_o,
output TagType tag_o,
output logic mask_o,
output AuxType aux_o,
// Output handshake
output logic out_valid_o,
input logic out_ready_i,
// Indication of valid data in flight
output logic busy_o,
// External register enable override
input logic [ExtRegEnaWidth-1:0] reg_ena_i
);
fpnew_fma #(
.FpFormat(FpFormat)
) top (
.clk_i(clk_i),
.rst_ni(rst_ni),
.operands_i(operands_i),
.is_boxed_i(is_boxed_i),
.rnd_mode_i(rnd_mode_i),
.op_i(op_i),
.op_mod_i(op_mod_i),
.tag_i(tag_i),
.mask_i(mask_i),
.aux_i(aux_i),
.in_valid_i(in_valid_i),
.in_ready_o(in_ready_o),
.flush_i(flush_i),
.result_o(result_o),
.status_o(status_o),
.extension_bit_o(extension_bit_o),
.tag_o(tag_o),
.mask_o(mask_o),
.aux_o(aux_o),
.out_valid_o(out_valid_o),
.out_ready_i(out_ready_i),
.busy_o(busy_o),
.reg_ena_i(reg_ena_i)
);
endmodule

69
fma_float32.sv
Unescape View File

@ -0,0 +1,69 @@
module top #(
parameter fpnew_pkg::fp_format_e FpFormat = fpnew_pkg::FP32,
parameter int unsigned NumPipeRegs = 0,
parameter fpnew_pkg::pipe_config_t PipeConfig = fpnew_pkg::BEFORE,
parameter type TagType = logic,
parameter type AuxType = logic,
localparam int unsigned WIDTH = 32,
localparam int unsigned ExtRegEnaWidth = 1
) (
input logic clk_i,
input logic rst_ni,
// Input signals
input logic [2:0][WIDTH-1:0] operands_i, // 3 operands
input logic [2:0] is_boxed_i, // 3 operands
input fpnew_pkg::roundmode_e rnd_mode_i,
input fpnew_pkg::operation_e op_i,
input logic op_mod_i,
input TagType tag_i,
input logic mask_i,
input AuxType aux_i,
// Input Handshake
input logic in_valid_i,
output logic in_ready_o,
input logic flush_i,
// Output signals
output logic [WIDTH-1:0] result_o,
output fpnew_pkg::status_t status_o,
output logic extension_bit_o,
output TagType tag_o,
output logic mask_o,
output AuxType aux_o,
// Output handshake
output logic out_valid_o,
input logic out_ready_i,
// Indication of valid data in flight
output logic busy_o,
// External register enable override
input logic [ExtRegEnaWidth-1:0] reg_ena_i
);
fpnew_fma #(
.FpFormat(FpFormat)
) top (
.clk_i(clk_i),
.rst_ni(rst_ni),
.operands_i(operands_i),
.is_boxed_i(is_boxed_i),
.rnd_mode_i(rnd_mode_i),
.op_i(op_i),
.op_mod_i(op_mod_i),
.tag_i(tag_i),
.mask_i(mask_i),
.aux_i(aux_i),
.in_valid_i(in_valid_i),
.in_ready_o(in_ready_o),
.flush_i(flush_i),
.result_o(result_o),
.status_o(status_o),
.extension_bit_o(extension_bit_o),
.tag_o(tag_o),
.mask_o(mask_o),
.aux_o(aux_o),
.out_valid_o(out_valid_o),
.out_ready_i(out_ready_i),
.busy_o(busy_o),
.reg_ena_i(reg_ena_i)
);
endmodule

69
fma_float64.sv
Unescape View File

@ -0,0 +1,69 @@
module top #(
parameter fpnew_pkg::fp_format_e FpFormat = fpnew_pkg::FP64,
parameter int unsigned NumPipeRegs = 0,
parameter fpnew_pkg::pipe_config_t PipeConfig = fpnew_pkg::BEFORE,
parameter type TagType = logic,
parameter type AuxType = logic,
localparam int unsigned WIDTH = 64,
localparam int unsigned ExtRegEnaWidth = 1
) (
input logic clk_i,
input logic rst_ni,
// Input signals
input logic [2:0][WIDTH-1:0] operands_i, // 3 operands
input logic [2:0] is_boxed_i, // 3 operands
input fpnew_pkg::roundmode_e rnd_mode_i,
input fpnew_pkg::operation_e op_i,
input logic op_mod_i,
input TagType tag_i,
input logic mask_i,
input AuxType aux_i,
// Input Handshake
input logic in_valid_i,
output logic in_ready_o,
input logic flush_i,
// Output signals
output logic [WIDTH-1:0] result_o,
output fpnew_pkg::status_t status_o,
output logic extension_bit_o,
output TagType tag_o,
output logic mask_o,
output AuxType aux_o,
// Output handshake
output logic out_valid_o,
input logic out_ready_i,
// Indication of valid data in flight
output logic busy_o,
// External register enable override
input logic [ExtRegEnaWidth-1:0] reg_ena_i
);
fpnew_fma #(
.FpFormat(FpFormat)
) top (
.clk_i(clk_i),
.rst_ni(rst_ni),
.operands_i(operands_i),
.is_boxed_i(is_boxed_i),
.rnd_mode_i(rnd_mode_i),
.op_i(op_i),
.op_mod_i(op_mod_i),
.tag_i(tag_i),
.mask_i(mask_i),
.aux_i(aux_i),
.in_valid_i(in_valid_i),
.in_ready_o(in_ready_o),
.flush_i(flush_i),
.result_o(result_o),
.status_o(status_o),
.extension_bit_o(extension_bit_o),
.tag_o(tag_o),
.mask_o(mask_o),
.aux_o(aux_o),
.out_valid_o(out_valid_o),
.out_ready_i(out_ready_i),
.busy_o(busy_o),
.reg_ena_i(reg_ena_i)
);
endmodule

6
main.py
Unescape View File

@ -0,0 +1,6 @@
def main():
print("Hello from mlhw!")
if __name__ == "__main__":
main()

7
pyproject.toml
Unescape View File

@ -0,0 +1,7 @@
[project]
name = "mlhw"
version = "0.1.0"
description = "Machine Learning Hardware Modules"
readme = "README.md"
requires-python = ">=3.14"
dependencies = []
Write Preview
Loading…
Cancel
Save