matthew1kalasky3
/
microwatt
forked from cores/microwatt
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity
You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
master
fpu-init
loadstore-init
core_debug-init
icache-unused-sig
icache-insn-u-state
dcache-unused-sig
unused-sig
divider-init
loadstore-pmu-init
icache-pmu-events
fpu-typo
less-fpga-init
caravel-mpw6-20220530
caravel-mpw5-20220323
caravel-mpw5-20220322
alt-reset-address
log2ceil-issue
fpu-constant
asic-3
boxarty-20211011
icbi-issue
orange-crab-freq
dcache-nc-fix
remove-potato-uart
cache-tlb-parameters-2
caravel-20210114
caravel-20210105
jtag-port-2
jtag-port
nia-debug
Branches Tags
${ item.name }
Create tag ${ searchTerm }
Create branch ${ searchTerm }
from '2224b28c2c'
${ noResults }
microwatt/xilinx-mult.vhdl

995 lines
28 KiB
VHDL
Raw Normal View History Unescape Escape

multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
library ieee;
use ieee.std_logic_1164.all;
use ieee.numeric_std.all;
library work;
use work.common.all;
library unisim;
use unisim.vcomponents.all;
entity multiply is
port (
clk : in std_logic;
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
m_in : in MultiplyInputType;
m_out : out MultiplyOutputType
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
);
end entity multiply;
architecture behaviour of multiply is
signal m00_p, m01_p, m02_p, m03_p : std_ulogic_vector(47 downto 0);
signal m00_pc : std_ulogic_vector(47 downto 0);
signal m10_p, m11_p, m12_p, m13_p : std_ulogic_vector(47 downto 0);
signal m11_pc, m12_pc, m13_pc : std_ulogic_vector(47 downto 0);
signal m20_p, m21_p, m22_p, m23_p : std_ulogic_vector(47 downto 0);
signal s0_pc, s1_pc : std_ulogic_vector(47 downto 0);
signal product : std_ulogic_vector(127 downto 0);
signal addend : std_ulogic_vector(127 downto 0);
signal s0_carry, p0_carry : std_ulogic_vector(3 downto 0);
signal p0_mask : std_ulogic_vector(47 downto 0);
signal p0_pat, p0_patb : std_ulogic;
signal p1_pat, p1_patb : std_ulogic;
signal req_32bit, r32_1 : std_ulogic;
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
signal rnot_1 : std_ulogic;
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
signal valid_1 : std_ulogic;
execute1: Take an extra cycle for OE=1 multiply instructions We now expect the overflow signal from the multiplier to come along one cycle later than the product. This breaks up a long combinatorial path and improves timing. This also changes some uses of v.<field> to r.<field> in the slow op logic, which should help timing as well. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
signal overflow, ovf_in : std_ulogic;
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
begin
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
addend <= m_in.addend;
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
m00: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
INMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
MREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
OPMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
PREG => 1
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
)
port map (
A => "0000000" & m_in.data1(22 downto 0),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(16 downto 0),
BCIN => (others => '0'),
C => "00000000000000" & addend(33 downto 0),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEC => '0',
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEM => '0',
CEP => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m00_p,
PCIN => (others => '0'),
PCOUT => m00_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m01: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "0000000" & m_in.data1(22 downto 0),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(33 downto 17),
BCIN => (others => '0'),
C => (others => '0'),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '1',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "1010101",
P => m01_p,
PCIN => m00_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m02: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
INMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
MREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
OPMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
PREG => 1
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
)
port map (
A => "0000000" & m_in.data1(22 downto 0),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(50 downto 34),
BCIN => (others => '0'),
C => x"0000000" & "000" & addend(50 downto 34),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEC => '0',
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEM => '0',
CEP => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m02_p,
PCIN => (others => '0'),
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m03: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
INMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
MREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
OPMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
PREG => 1
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
)
port map (
A => "0000000" & m_in.data1(22 downto 0),
ACIN => (others => '0'),
ALUMODE => "0000",
B => "00000" & m_in.data2(63 downto 51),
BCIN => (others => '0'),
C => x"000000" & '0' & addend(73 downto 51),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEC => '0',
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEM => '0',
CEP => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m03_p,
PCIN => (others => '0'),
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m10: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
CREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "0000000000000" & m_in.data1(39 downto 23),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(16 downto 0),
BCIN => (others => '0'),
C => x"000" & "00" & m01_p(39 downto 6),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '0',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m10_p,
PCIN => (others => '0'),
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m11: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
CREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "0000000000000" & m_in.data1(39 downto 23),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(33 downto 17),
BCIN => (others => '0'),
C => x"000" & "00" & m02_p(39 downto 6),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '0',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m11_p,
PCIN => (others => '0'),
PCOUT => m11_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m12: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
CREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "0000000000000" & m_in.data1(39 downto 23),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(50 downto 34),
BCIN => (others => '0'),
C => x"0000" & '0' & m03_p(36 downto 6),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '0',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m12_p,
PCIN => (others => '0'),
PCOUT => m12_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m13: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "0000000000000" & m_in.data1(39 downto 23),
ACIN => (others => '0'),
ALUMODE => "0000",
B => "00000" & m_in.data2(63 downto 51),
BCIN => (others => '0'),
C => x"0000000" & "000" & addend(90 downto 74),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '1',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m13_p,
PCIN => (others => '0'),
PCOUT => m13_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m20: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "000000" & m_in.data1(63 downto 40),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(16 downto 0),
BCIN => (others => '0'),
C => (others => '0'),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '1',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0010101",
P => m20_p,
PCIN => m11_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m21: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "000000" & m_in.data1(63 downto 40),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(33 downto 17),
BCIN => (others => '0'),
C => (others => '0'),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '1',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0010101",
P => m21_p,
PCIN => m12_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m22: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "000000" & m_in.data1(63 downto 40),
ACIN => (others => '0'),
ALUMODE => "0000",
B => '0' & m_in.data2(50 downto 34),
BCIN => (others => '0'),
C => (others => '0'),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '1',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0010101",
P => m22_p,
PCIN => m13_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
m23: DSP48E1
generic map (
ACASCREG => 0,
ALUMODEREG => 0,
AREG => 0,
BCASCREG => 0,
BREG => 0,
CARRYINREG => 0,
CARRYINSELREG => 0,
INMODEREG => 0,
OPMODEREG => 0,
PREG => 0
)
port map (
A => "000000" & m_in.data1(63 downto 40),
ACIN => (others => '0'),
ALUMODE => "0000",
B => "00000" & m_in.data2(63 downto 51),
BCIN => (others => '0'),
C => x"00" & "000" & addend(127 downto 91),
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CEA1 => '0',
CEA2 => '0',
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
CEB2 => '0',
CEC => '1',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEM => m_in.valid,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0110101",
P => m23_p,
PCIN => (others => '0'),
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
s0: DSP48E1
generic map (
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
ACASCREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
ALUMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
AREG => 0,
BCASCREG => 0,
BREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CARRYINREG => 0,
CARRYINSELREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CREG => 0,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
INMODEREG => 0,
MREG => 0,
OPMODEREG => 0,
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
PREG => 1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
USE_MULT => "none"
)
port map (
A => m22_p(5 downto 0) & x"0000" & m10_p(34 downto 27),
ACIN => (others => '0'),
ALUMODE => "0000",
B => m10_p(26 downto 9),
BCIN => (others => '0'),
C => m20_p(39 downto 0) & m02_p(5 downto 0) & "00",
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CARRYOUT => s0_carry,
CEA1 => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEA2 => '0',
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEB2 => '0',
CEC => '0',
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
CEM => '0',
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
CEP => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0001111",
PCIN => (others => '0'),
PCOUT => s0_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
s1: DSP48E1
generic map (
ACASCREG => 1,
ALUMODEREG => 0,
AREG => 1,
BCASCREG => 1,
BREG => 1,
CARRYINREG => 0,
CARRYINSELREG => 0,
CREG => 1,
INMODEREG => 0,
MREG => 0,
OPMODEREG => 0,
PREG => 0,
USE_MULT => "none"
)
port map (
A => x"000" & m22_p(41 downto 24),
ACIN => (others => '0'),
ALUMODE => "0000",
B => m22_p(23 downto 6),
BCIN => (others => '0'),
C => m23_p(36 downto 0) & x"00" & "0" & m20_p(41 downto 40),
CARRYCASCIN => '0',
CARRYIN => s0_carry(3),
CARRYINSEL => "000",
CEA1 => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEA2 => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEAD => '0',
CEALUMODE => '0',
CEB1 => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEB2 => valid_1,
CEC => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
CEM => '0',
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0001111",
PCIN => (others => '0'),
PCOUT => s1_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
-- mask is 0 for 32-bit ops, 0x0000ffffffff for 64-bit
p0_mask(47 downto 31) <= (others => '0');
p0_mask(30 downto 0) <= (others => not r32_1);
p0: DSP48E1
generic map (
ACASCREG => 1,
ALUMODEREG => 1,
AREG => 1,
BCASCREG => 1,
BREG => 1,
CARRYINREG => 0,
CARRYINSELREG => 0,
CREG => 1,
INMODEREG => 0,
MREG => 0,
OPMODEREG => 0,
PREG => 0,
SEL_MASK => "C",
USE_MULT => "none",
USE_PATTERN_DETECT => "PATDET"
)
port map (
A => m21_p(22 downto 0) & m03_p(5 downto 0) & '0',
ACIN => (others => '0'),
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
ALUMODE => "00" & rnot_1 & '0',
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
B => (others => '0'),
BCIN => (others => '0'),
C => p0_mask,
CARRYCASCIN => '0',
CARRYIN => '0',
CARRYINSEL => "000",
CARRYOUT => p0_carry,
CEA1 => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEA2 => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEAD => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEALUMODE => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEB1 => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEB2 => valid_1,
CEC => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
CEM => '0',
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0010011",
P => product(79 downto 32),
PATTERNDETECT => p0_pat,
PATTERNBDETECT => p0_patb,
PCIN => s0_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
p1: DSP48E1
generic map (
ACASCREG => 1,
ALUMODEREG => 1,
AREG => 1,
BCASCREG => 1,
BREG => 1,
CARRYINREG => 0,
CARRYINSELREG => 0,
CREG => 0,
INMODEREG => 0,
MASK => x"000000000000",
MREG => 0,
OPMODEREG => 0,
PREG => 0,
USE_MULT => "none",
USE_PATTERN_DETECT => "PATDET"
)
port map (
A => x"0000000" & '0' & m21_p(41),
ACIN => (others => '0'),
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
ALUMODE => "00" & rnot_1 & '0',
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
B => m21_p(40 downto 23),
BCIN => (others => '0'),
C => (others => '0'),
CARRYCASCIN => '0',
CARRYIN => p0_carry(3),
CARRYINSEL => "000",
CEA1 => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEA2 => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEAD => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEALUMODE => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEB1 => '0',
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
CEB2 => valid_1,
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
CEC => '0',
CECARRYIN => '0',
CECTRL => '0',
CED => '0',
CEINMODE => '0',
CEM => '0',
CEP => '0',
CLK => clk,
D => (others => '0'),
INMODE => "00000",
MULTSIGNIN => '0',
OPMODE => "0010011",
P => product(127 downto 80),
PATTERNDETECT => p1_pat,
PATTERNBDETECT => p1_patb,
PCIN => s1_pc,
RSTA => '0',
RSTALLCARRYIN => '0',
RSTALUMODE => '0',
RSTB => '0',
RSTC => '0',
RSTCTRL => '0',
RSTD => '0',
RSTINMODE => '0',
RSTM => '0',
RSTP => '0'
);
mult_out: process(all)
variable ov : std_ulogic;
begin
-- set overflow if the high bits are neither all zeroes nor all ones
if req_32bit = '0' then
ov := not ((p1_pat and p0_pat) or (p1_patb and p0_patb));
else
ov := not ((p1_pat and p0_pat and not product(31)) or
(p1_patb and p0_patb and product(31)));
end if;
execute1: Take an extra cycle for OE=1 multiply instructions We now expect the overflow signal from the multiplier to come along one cycle later than the product. This breaks up a long combinatorial path and improves timing. This also changes some uses of v.<field> to r.<field> in the slow op logic, which should help timing as well. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
ovf_in <= ov;
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
m_out.result <= product;
execute1: Take an extra cycle for OE=1 multiply instructions We now expect the overflow signal from the multiplier to come along one cycle later than the product. This breaks up a long combinatorial path and improves timing. This also changes some uses of v.<field> to r.<field> in the slow op logic, which should help timing as well. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
m_out.overflow <= overflow;
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
end process;
process(clk)
begin
if rising_edge(clk) then
xilinx-mult: Move some registers later in the data flow This changes s0 to use the P register rather than the A/B/C input registers, thus improving the timing of the multiplier output. The m00, m02 and m03 multipliers now use their P registers rather than the M registers, moving the addition they do from the second cycle to the first. Also, the XOR that inverts the 32 LSBs is moved before the output register. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
3 years ago
if rnot_1 = '0' then
product(31 downto 0) <= m10_p(8 downto 0) & m01_p(5 downto 0) & m00_p(16 downto 0);
else
product(31 downto 0) <= not (m10_p(8 downto 0) & m01_p(5 downto 0) & m00_p(16 downto 0));
end if;
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
m_out.valid <= valid_1;
valid_1 <= m_in.valid;
req_32bit <= r32_1;
r32_1 <= m_in.is_32bit;
multiplier: Generalize interface to the multiplier This makes the interface to the multiplier more general so an instance of it can be used in the FPU. It now has a 128-bit addend that is added on to the product. Instead of an input to negate the output, it now has a "not_result" input to complement the output. Execute1 uses not_result=1 and addend=-1 to get the effect of negating the output. The interface is defined this way because this is what can be done easily with the Xilinx DSP slices in xilinx-mult.vhdl. This also adds clock enable signals to the DSP slices, mostly for the sake of reducing power consumption. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
rnot_1 <= m_in.not_result;
execute1: Take an extra cycle for OE=1 multiply instructions We now expect the overflow signal from the multiplier to come along one cycle later than the product. This breaks up a long combinatorial path and improves timing. This also changes some uses of v.<field> to r.<field> in the slow op logic, which should help timing as well. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
overflow <= ovf_in;
multiply: Use DSP48 slices for multiplication on Xilinx FPGAs This adds a custom implementation of the multiplier which uses 16 DSP48E1 slices to do a 64x64 bit multiplication in 2 cycles. Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
5 years ago
end if;
end process;
end architecture behaviour;