|
|
|
library ieee;
|
|
|
|
use ieee.std_logic_1164.all;
|
|
|
|
use ieee.numeric_std.all;
|
|
|
|
|
|
|
|
library work;
|
|
|
|
use work.decode_types.all;
|
|
|
|
use work.common.all;
|
|
|
|
use work.helpers.all;
|
|
|
|
use work.insn_helpers.all;
|
|
|
|
|
|
|
|
entity decode2 is
|
|
|
|
generic (
|
|
|
|
EX1_BYPASS : boolean := true;
|
|
|
|
HAS_FPU : boolean := true;
|
|
|
|
-- Non-zero to enable log data collection
|
|
|
|
LOG_LENGTH : natural := 0
|
|
|
|
);
|
|
|
|
port (
|
|
|
|
clk : in std_ulogic;
|
|
|
|
rst : in std_ulogic;
|
|
|
|
|
|
|
|
complete_in : in instr_tag_t;
|
|
|
|
busy_in : in std_ulogic;
|
|
|
|
stall_out : out std_ulogic;
|
|
|
|
|
|
|
|
stopped_out : out std_ulogic;
|
|
|
|
|
|
|
|
flush_in: in std_ulogic;
|
|
|
|
|
|
|
|
d_in : in Decode1ToDecode2Type;
|
|
|
|
|
|
|
|
e_out : out Decode2ToExecute1Type;
|
|
|
|
|
|
|
|
r_in : in RegisterFileToDecode2Type;
|
|
|
|
r_out : out Decode2ToRegisterFileType;
|
|
|
|
|
|
|
|
c_in : in CrFileToDecode2Type;
|
|
|
|
c_out : out Decode2ToCrFileType;
|
|
|
|
|
|
|
|
execute_bypass : in bypass_data_t;
|
|
|
|
execute_cr_bypass : in cr_bypass_data_t;
|
|
|
|
execute2_bypass : in bypass_data_t;
|
|
|
|
execute2_cr_bypass : in cr_bypass_data_t;
|
|
|
|
|
|
|
|
log_out : out std_ulogic_vector(9 downto 0)
|
|
|
|
);
|
|
|
|
end entity decode2;
|
|
|
|
|
|
|
|
architecture behaviour of decode2 is
|
|
|
|
type reg_type is record
|
|
|
|
e : Decode2ToExecute1Type;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
repeat : std_ulogic;
|
|
|
|
end record;
|
|
|
|
|
|
|
|
signal r, rin : reg_type;
|
|
|
|
|
|
|
|
signal deferred : std_ulogic;
|
|
|
|
|
|
|
|
type decode_input_reg_t is record
|
|
|
|
reg_valid : std_ulogic;
|
|
|
|
reg : gspr_index_t;
|
|
|
|
data : std_ulogic_vector(63 downto 0);
|
|
|
|
end record;
|
|
|
|
|
|
|
|
type decode_output_reg_t is record
|
|
|
|
reg_valid : std_ulogic;
|
|
|
|
reg : gspr_index_t;
|
|
|
|
end record;
|
|
|
|
|
|
|
|
function decode_input_reg_a (t : input_reg_a_t; insn_in : std_ulogic_vector(31 downto 0);
|
|
|
|
reg_data : std_ulogic_vector(63 downto 0);
|
|
|
|
ispr : gspr_index_t;
|
|
|
|
instr_addr : std_ulogic_vector(63 downto 0))
|
|
|
|
return decode_input_reg_t is
|
|
|
|
begin
|
|
|
|
if t = RA or (t = RA_OR_ZERO and insn_ra(insn_in) /= "00000") then
|
|
|
|
return ('1', gpr_to_gspr(insn_ra(insn_in)), reg_data);
|
|
|
|
elsif t = SPR then
|
|
|
|
-- ISPR must be either a valid fast SPR number or all 0 for a slow SPR.
|
|
|
|
-- If it's all 0, we don't treat it as a dependency as slow SPRs
|
|
|
|
-- operations are single issue.
|
|
|
|
--
|
|
|
|
assert is_fast_spr(ispr) = '1' or ispr = "0000000"
|
|
|
|
report "Decode A says SPR but ISPR is invalid:" &
|
|
|
|
to_hstring(ispr) severity failure;
|
|
|
|
return (is_fast_spr(ispr), ispr, reg_data);
|
|
|
|
elsif t = CIA then
|
|
|
|
return ('0', (others => '0'), instr_addr);
|
|
|
|
elsif HAS_FPU and t = FRA then
|
|
|
|
return ('1', fpr_to_gspr(insn_fra(insn_in)), reg_data);
|
|
|
|
else
|
|
|
|
return ('0', (others => '0'), (others => '0'));
|
|
|
|
end if;
|
|
|
|
end;
|
|
|
|
|
|
|
|
function decode_input_reg_b (t : input_reg_b_t; insn_in : std_ulogic_vector(31 downto 0);
|
|
|
|
reg_data : std_ulogic_vector(63 downto 0);
|
|
|
|
ispr : gspr_index_t) return decode_input_reg_t is
|
|
|
|
variable ret : decode_input_reg_t;
|
|
|
|
begin
|
|
|
|
case t is
|
|
|
|
when RB =>
|
|
|
|
ret := ('1', gpr_to_gspr(insn_rb(insn_in)), reg_data);
|
|
|
|
when FRB =>
|
|
|
|
if HAS_FPU then
|
|
|
|
ret := ('1', fpr_to_gspr(insn_frb(insn_in)), reg_data);
|
|
|
|
else
|
|
|
|
ret := ('0', (others => '0'), (others => '0'));
|
|
|
|
end if;
|
|
|
|
when CONST_UI =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(unsigned(insn_ui(insn_in)), 64)));
|
|
|
|
when CONST_SI =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(signed(insn_si(insn_in)), 64)));
|
|
|
|
when CONST_SI_HI =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(signed(insn_si(insn_in)) & x"0000", 64)));
|
|
|
|
when CONST_UI_HI =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(unsigned(insn_si(insn_in)) & x"0000", 64)));
|
|
|
|
when CONST_LI =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(signed(insn_li(insn_in)) & "00", 64)));
|
|
|
|
when CONST_BD =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(signed(insn_bd(insn_in)) & "00", 64)));
|
|
|
|
when CONST_DS =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(signed(insn_ds(insn_in)) & "00", 64)));
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
when CONST_DQ =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(signed(insn_dq(insn_in)) & "0000", 64)));
|
|
|
|
when CONST_DXHI4 =>
|
|
|
|
ret := ('0', (others => '0'), std_ulogic_vector(resize(signed(insn_dx(insn_in)) & x"0004", 64)));
|
|
|
|
when CONST_M1 =>
|
|
|
|
ret := ('0', (others => '0'), x"FFFFFFFFFFFFFFFF");
|
|
|
|
when CONST_SH =>
|
|
|
|
ret := ('0', (others => '0'), x"00000000000000" & "00" & insn_in(1) & insn_in(15 downto 11));
|
|
|
|
when CONST_SH32 =>
|
|
|
|
ret := ('0', (others => '0'), x"00000000000000" & "000" & insn_in(15 downto 11));
|
|
|
|
when SPR =>
|
|
|
|
-- ISPR must be either a valid fast SPR number or all 0 for a slow SPR.
|
|
|
|
-- If it's all 0, we don't treat it as a dependency as slow SPRs
|
|
|
|
-- operations are single issue.
|
|
|
|
assert is_fast_spr(ispr) = '1' or ispr = "0000000"
|
|
|
|
report "Decode B says SPR but ISPR is invalid:" &
|
|
|
|
to_hstring(ispr) severity failure;
|
|
|
|
ret := (is_fast_spr(ispr), ispr, reg_data);
|
|
|
|
when NONE =>
|
|
|
|
ret := ('0', (others => '0'), (others => '0'));
|
|
|
|
end case;
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
end;
|
|
|
|
|
|
|
|
function decode_input_reg_c (t : input_reg_c_t; insn_in : std_ulogic_vector(31 downto 0);
|
|
|
|
reg_data : std_ulogic_vector(63 downto 0)) return decode_input_reg_t is
|
|
|
|
begin
|
|
|
|
case t is
|
|
|
|
when RS =>
|
|
|
|
return ('1', gpr_to_gspr(insn_rs(insn_in)), reg_data);
|
|
|
|
when RCR =>
|
|
|
|
return ('1', gpr_to_gspr(insn_rcreg(insn_in)), reg_data);
|
|
|
|
when FRS =>
|
|
|
|
if HAS_FPU then
|
|
|
|
return ('1', fpr_to_gspr(insn_frt(insn_in)), reg_data);
|
|
|
|
else
|
|
|
|
return ('0', (others => '0'), (others => '0'));
|
|
|
|
end if;
|
|
|
|
when FRC =>
|
|
|
|
if HAS_FPU then
|
|
|
|
return ('1', fpr_to_gspr(insn_frc(insn_in)), reg_data);
|
|
|
|
else
|
|
|
|
return ('0', (others => '0'), (others => '0'));
|
|
|
|
end if;
|
|
|
|
when NONE =>
|
|
|
|
return ('0', (others => '0'), (others => '0'));
|
|
|
|
end case;
|
|
|
|
end;
|
|
|
|
|
|
|
|
function decode_output_reg (t : output_reg_a_t; insn_in : std_ulogic_vector(31 downto 0);
|
|
|
|
ispr : gspr_index_t) return decode_output_reg_t is
|
|
|
|
begin
|
|
|
|
case t is
|
|
|
|
when RT =>
|
|
|
|
return ('1', gpr_to_gspr(insn_rt(insn_in)));
|
|
|
|
when RA =>
|
|
|
|
return ('1', gpr_to_gspr(insn_ra(insn_in)));
|
|
|
|
when FRT =>
|
|
|
|
if HAS_FPU then
|
|
|
|
return ('1', fpr_to_gspr(insn_frt(insn_in)));
|
|
|
|
else
|
|
|
|
return ('0', "0000000");
|
|
|
|
end if;
|
|
|
|
when SPR =>
|
|
|
|
-- ISPR must be either a valid fast SPR number or all 0 for a slow SPR.
|
|
|
|
-- If it's all 0, we don't treat it as a dependency as slow SPRs
|
|
|
|
-- operations are single issue.
|
|
|
|
assert is_fast_spr(ispr) = '1' or ispr = "0000000"
|
|
|
|
report "Decode B says SPR but ISPR is invalid:" &
|
|
|
|
to_hstring(ispr) severity failure;
|
|
|
|
return (is_fast_spr(ispr), ispr);
|
|
|
|
when NONE =>
|
|
|
|
return ('0', "0000000");
|
|
|
|
end case;
|
|
|
|
end;
|
|
|
|
|
|
|
|
function decode_rc (t : rc_t; insn_in : std_ulogic_vector(31 downto 0)) return std_ulogic is
|
|
|
|
begin
|
|
|
|
case t is
|
|
|
|
when RC =>
|
|
|
|
return insn_rc(insn_in);
|
|
|
|
when ONE =>
|
|
|
|
return '1';
|
|
|
|
when NONE =>
|
|
|
|
return '0';
|
|
|
|
end case;
|
|
|
|
end;
|
|
|
|
|
|
|
|
-- control signals that are derived from insn_type
|
|
|
|
type mux_select_array_t is array(insn_type_t) of std_ulogic_vector(2 downto 0);
|
|
|
|
|
|
|
|
constant result_select : mux_select_array_t := (
|
|
|
|
OP_AND => "001", -- logical_result
|
|
|
|
OP_OR => "001",
|
|
|
|
OP_XOR => "001",
|
|
|
|
OP_PRTY => "001",
|
|
|
|
OP_CMPB => "001",
|
|
|
|
OP_EXTS => "001",
|
|
|
|
OP_BPERM => "001",
|
|
|
|
OP_BCD => "001",
|
|
|
|
OP_MTSPR => "001",
|
|
|
|
OP_RLC => "010", -- rotator_result
|
|
|
|
OP_RLCL => "010",
|
|
|
|
OP_RLCR => "010",
|
|
|
|
OP_SHL => "010",
|
|
|
|
OP_SHR => "010",
|
|
|
|
OP_EXTSWSLI => "010",
|
|
|
|
OP_MUL_L64 => "011", -- muldiv_result
|
|
|
|
OP_B => "110", -- next_nia
|
|
|
|
OP_BC => "110",
|
|
|
|
OP_BCREG => "110",
|
|
|
|
OP_ADDG6S => "111", -- misc_result
|
|
|
|
OP_ISEL => "111",
|
|
|
|
OP_DARN => "111",
|
|
|
|
OP_MFMSR => "111",
|
|
|
|
OP_MFCR => "111",
|
|
|
|
OP_SETB => "111",
|
|
|
|
others => "000" -- default to adder_result
|
|
|
|
);
|
|
|
|
|
|
|
|
constant subresult_select : mux_select_array_t := (
|
|
|
|
OP_MUL_L64 => "000", -- muldiv_result
|
|
|
|
OP_MUL_H64 => "001",
|
|
|
|
OP_MUL_H32 => "010",
|
|
|
|
OP_DIV => "011",
|
|
|
|
OP_DIVE => "011",
|
|
|
|
OP_MOD => "011",
|
|
|
|
OP_ADDG6S => "001", -- misc_result
|
|
|
|
OP_ISEL => "010",
|
|
|
|
OP_DARN => "011",
|
|
|
|
OP_MFMSR => "100",
|
|
|
|
OP_MFCR => "101",
|
|
|
|
OP_SETB => "110",
|
|
|
|
OP_CMP => "000", -- cr_result
|
|
|
|
OP_CMPRB => "001",
|
|
|
|
OP_CMPEQB => "010",
|
|
|
|
OP_CROP => "011",
|
|
|
|
OP_MCRXRX => "100",
|
|
|
|
OP_MTCRF => "101",
|
|
|
|
others => "000"
|
|
|
|
);
|
|
|
|
|
|
|
|
-- issue control signals
|
|
|
|
signal control_valid_in : std_ulogic;
|
|
|
|
signal control_valid_out : std_ulogic;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
signal control_stall_out : std_ulogic;
|
|
|
|
signal control_sgl_pipe : std_logic;
|
|
|
|
|
|
|
|
signal gpr_write_valid : std_ulogic;
|
|
|
|
signal gpr_write : gspr_index_t;
|
|
|
|
|
|
|
|
signal gpr_a_read_valid : std_ulogic;
|
|
|
|
signal gpr_a_read : gspr_index_t;
|
|
|
|
signal gpr_a_bypass : std_ulogic_vector(1 downto 0);
|
|
|
|
|
|
|
|
signal gpr_b_read_valid : std_ulogic;
|
|
|
|
signal gpr_b_read : gspr_index_t;
|
|
|
|
signal gpr_b_bypass : std_ulogic_vector(1 downto 0);
|
|
|
|
|
|
|
|
signal gpr_c_read_valid : std_ulogic;
|
|
|
|
signal gpr_c_read : gspr_index_t;
|
|
|
|
signal gpr_c_bypass : std_ulogic_vector(1 downto 0);
|
|
|
|
|
|
|
|
signal cr_read_valid : std_ulogic;
|
|
|
|
signal cr_write_valid : std_ulogic;
|
|
|
|
signal cr_bypass : std_ulogic_vector(1 downto 0);
|
|
|
|
|
|
|
|
signal instr_tag : instr_tag_t;
|
|
|
|
|
|
|
|
begin
|
|
|
|
control_0: entity work.control
|
|
|
|
generic map (
|
|
|
|
EX1_BYPASS => EX1_BYPASS
|
|
|
|
)
|
|
|
|
port map (
|
|
|
|
clk => clk,
|
|
|
|
rst => rst,
|
|
|
|
|
|
|
|
complete_in => complete_in,
|
|
|
|
valid_in => control_valid_in,
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
repeated => r.repeat,
|
|
|
|
busy_in => busy_in,
|
|
|
|
deferred => deferred,
|
|
|
|
flush_in => flush_in,
|
|
|
|
sgl_pipe_in => control_sgl_pipe,
|
|
|
|
stop_mark_in => d_in.stop_mark,
|
|
|
|
|
|
|
|
gpr_write_valid_in => gpr_write_valid,
|
|
|
|
gpr_write_in => gpr_write,
|
|
|
|
|
|
|
|
gpr_a_read_valid_in => gpr_a_read_valid,
|
|
|
|
gpr_a_read_in => gpr_a_read,
|
|
|
|
|
|
|
|
gpr_b_read_valid_in => gpr_b_read_valid,
|
|
|
|
gpr_b_read_in => gpr_b_read,
|
|
|
|
|
|
|
|
gpr_c_read_valid_in => gpr_c_read_valid,
|
|
|
|
gpr_c_read_in => gpr_c_read,
|
|
|
|
|
|
|
|
execute_next_tag => execute_bypass.tag,
|
|
|
|
execute_next_cr_tag => execute_cr_bypass.tag,
|
|
|
|
execute2_next_tag => execute2_bypass.tag,
|
|
|
|
execute2_next_cr_tag => execute2_cr_bypass.tag,
|
|
|
|
|
|
|
|
cr_read_in => cr_read_valid,
|
|
|
|
cr_write_in => cr_write_valid,
|
|
|
|
cr_bypass => cr_bypass,
|
|
|
|
|
|
|
|
valid_out => control_valid_out,
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
stall_out => control_stall_out,
|
|
|
|
stopped_out => stopped_out,
|
|
|
|
|
|
|
|
gpr_bypass_a => gpr_a_bypass,
|
|
|
|
gpr_bypass_b => gpr_b_bypass,
|
|
|
|
gpr_bypass_c => gpr_c_bypass,
|
|
|
|
|
|
|
|
instr_tag_out => instr_tag
|
|
|
|
);
|
|
|
|
|
|
|
|
deferred <= r.e.valid and busy_in;
|
|
|
|
|
|
|
|
decode2_0: process(clk)
|
|
|
|
begin
|
|
|
|
if rising_edge(clk) then
|
|
|
|
if rst = '1' or flush_in = '1' or deferred = '0' then
|
|
|
|
if rin.e.valid = '1' then
|
|
|
|
report "execute " & to_hstring(rin.e.nia);
|
|
|
|
end if;
|
|
|
|
r <= rin;
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
end process;
|
|
|
|
|
|
|
|
c_out.read <= d_in.decode.input_cr;
|
|
|
|
|
|
|
|
decode2_1: process(all)
|
|
|
|
variable v : reg_type;
|
|
|
|
variable mul_a : std_ulogic_vector(63 downto 0);
|
|
|
|
variable mul_b : std_ulogic_vector(63 downto 0);
|
|
|
|
variable decoded_reg_a : decode_input_reg_t;
|
|
|
|
variable decoded_reg_b : decode_input_reg_t;
|
|
|
|
variable decoded_reg_c : decode_input_reg_t;
|
|
|
|
variable decoded_reg_o : decode_output_reg_t;
|
|
|
|
variable length : std_ulogic_vector(3 downto 0);
|
|
|
|
variable op : insn_type_t;
|
|
|
|
begin
|
|
|
|
v := r;
|
|
|
|
|
|
|
|
v.e := Decode2ToExecute1Init;
|
|
|
|
|
|
|
|
mul_a := (others => '0');
|
|
|
|
mul_b := (others => '0');
|
|
|
|
|
|
|
|
--v.e.input_cr := d_in.decode.input_cr;
|
|
|
|
v.e.output_cr := d_in.decode.output_cr;
|
|
|
|
|
|
|
|
-- Work out whether XER common bits are set
|
|
|
|
v.e.output_xer := d_in.decode.output_carry;
|
|
|
|
case d_in.decode.insn_type is
|
|
|
|
when OP_ADD | OP_MUL_L64 | OP_DIV | OP_DIVE =>
|
|
|
|
-- OE field is valid in OP_ADD/OP_MUL_L64 with major opcode 31 only
|
|
|
|
if d_in.insn(31 downto 26) = "011111" and insn_oe(d_in.insn) = '1' then
|
|
|
|
v.e.oe := '1';
|
|
|
|
v.e.output_xer := '1';
|
|
|
|
end if;
|
|
|
|
when OP_MTSPR =>
|
|
|
|
if decode_spr_num(d_in.insn) = SPR_XER then
|
|
|
|
v.e.output_xer := '1';
|
|
|
|
end if;
|
|
|
|
when others =>
|
|
|
|
end case;
|
|
|
|
|
|
|
|
decoded_reg_a := decode_input_reg_a (d_in.decode.input_reg_a, d_in.insn, r_in.read1_data, d_in.ispr1,
|
|
|
|
d_in.nia);
|
|
|
|
decoded_reg_b := decode_input_reg_b (d_in.decode.input_reg_b, d_in.insn, r_in.read2_data, d_in.ispr2);
|
|
|
|
decoded_reg_c := decode_input_reg_c (d_in.decode.input_reg_c, d_in.insn, r_in.read3_data);
|
|
|
|
decoded_reg_o := decode_output_reg (d_in.decode.output_reg_a, d_in.insn, d_in.ispro);
|
|
|
|
|
|
|
|
if d_in.decode.lr = '1' then
|
|
|
|
v.e.lr := insn_lk(d_in.insn);
|
|
|
|
-- b and bc have even major opcodes; bcreg is considered absolute
|
|
|
|
v.e.br_abs := insn_aa(d_in.insn) or d_in.insn(26);
|
|
|
|
end if;
|
|
|
|
op := d_in.decode.insn_type;
|
|
|
|
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
if d_in.decode.repeat /= NONE then
|
|
|
|
v.e.repeat := '1';
|
|
|
|
v.e.second := r.repeat;
|
|
|
|
case d_in.decode.repeat is
|
|
|
|
when DRSE =>
|
|
|
|
-- do RS|1,RS for LE; RS,RS|1 for BE
|
|
|
|
if r.repeat = d_in.big_endian then
|
|
|
|
decoded_reg_c.reg(0) := '1';
|
|
|
|
end if;
|
|
|
|
when DRTE =>
|
|
|
|
-- do RT|1,RT for LE; RT,RT|1 for BE
|
|
|
|
if r.repeat = d_in.big_endian then
|
|
|
|
decoded_reg_o.reg(0) := '1';
|
|
|
|
end if;
|
|
|
|
when DUPD =>
|
|
|
|
-- update-form loads, 2nd instruction writes RA
|
|
|
|
if r.repeat = '1' then
|
|
|
|
decoded_reg_o.reg := decoded_reg_a.reg;
|
|
|
|
end if;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
when others =>
|
|
|
|
end case;
|
|
|
|
elsif v.e.lr = '1' and decoded_reg_a.reg_valid = '1' then
|
|
|
|
-- bcl/bclrl/bctarl that needs to write both CTR and LR has to be doubled
|
|
|
|
v.e.repeat := '1';
|
|
|
|
v.e.second := r.repeat;
|
|
|
|
-- first one does CTR, second does LR
|
|
|
|
decoded_reg_o.reg(0) := not r.repeat;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
end if;
|
|
|
|
|
|
|
|
v.e.spr_select := d_in.spr_info;
|
|
|
|
|
|
|
|
r_out.read1_enable <= decoded_reg_a.reg_valid and d_in.valid;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
r_out.read1_reg <= decoded_reg_a.reg;
|
|
|
|
r_out.read2_enable <= decoded_reg_b.reg_valid and d_in.valid;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
r_out.read2_reg <= decoded_reg_b.reg;
|
|
|
|
r_out.read3_enable <= decoded_reg_c.reg_valid and d_in.valid;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
r_out.read3_reg <= decoded_reg_c.reg;
|
|
|
|
|
|
|
|
case d_in.decode.length is
|
|
|
|
when is1B =>
|
|
|
|
length := "0001";
|
|
|
|
when is2B =>
|
|
|
|
length := "0010";
|
|
|
|
when is4B =>
|
|
|
|
length := "0100";
|
|
|
|
when is8B =>
|
|
|
|
length := "1000";
|
|
|
|
when NONE =>
|
|
|
|
length := "0000";
|
|
|
|
end case;
|
|
|
|
|
|
|
|
-- execute unit
|
|
|
|
v.e.nia := d_in.nia;
|
|
|
|
v.e.unit := d_in.decode.unit;
|
|
|
|
v.e.fac := d_in.decode.facility;
|
|
|
|
v.e.instr_tag := instr_tag;
|
|
|
|
v.e.read_reg1 := decoded_reg_a.reg;
|
|
|
|
v.e.read_reg2 := decoded_reg_b.reg;
|
|
|
|
v.e.write_reg := decoded_reg_o.reg;
|
|
|
|
v.e.write_reg_enable := decoded_reg_o.reg_valid;
|
|
|
|
v.e.rc := decode_rc(d_in.decode.rc, d_in.insn);
|
|
|
|
v.e.xerc := c_in.read_xerc_data;
|
|
|
|
v.e.invert_a := d_in.decode.invert_a;
|
|
|
|
v.e.addm1 := '0';
|
|
|
|
v.e.insn_type := op;
|
|
|
|
v.e.invert_out := d_in.decode.invert_out;
|
|
|
|
v.e.input_carry := d_in.decode.input_carry;
|
|
|
|
v.e.output_carry := d_in.decode.output_carry;
|
|
|
|
v.e.is_32bit := d_in.decode.is_32bit;
|
|
|
|
v.e.is_signed := d_in.decode.is_signed;
|
|
|
|
v.e.insn := d_in.insn;
|
|
|
|
v.e.data_len := length;
|
|
|
|
v.e.byte_reverse := d_in.decode.byte_reverse;
|
|
|
|
v.e.sign_extend := d_in.decode.sign_extend;
|
|
|
|
v.e.update := d_in.decode.update;
|
|
|
|
v.e.reserve := d_in.decode.reserve;
|
|
|
|
v.e.br_pred := d_in.br_pred;
|
|
|
|
v.e.result_sel := result_select(op);
|
|
|
|
v.e.sub_select := subresult_select(op);
|
|
|
|
if op = OP_BC or op = OP_BCREG then
|
|
|
|
if d_in.insn(23) = '0' and r.repeat = '0' and
|
|
|
|
not (d_in.decode.insn_type = OP_BCREG and d_in.insn(10) = '0') then
|
|
|
|
-- decrement CTR if BO(2) = 0 and not bcctr
|
|
|
|
v.e.addm1 := '1';
|
|
|
|
v.e.result_sel := "000"; -- select adder output
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
if op = OP_MFSPR then
|
|
|
|
if is_fast_spr(d_in.ispr1) = '1' then
|
|
|
|
v.e.result_sel := "000"; -- adder_result, effectively a_in
|
|
|
|
elsif d_in.spr_info.valid = '0' then
|
|
|
|
-- Privileged mfspr to invalid/unimplemented SPR numbers
|
|
|
|
-- writes the contents of RT back to RT (i.e. it's a no-op)
|
|
|
|
v.e.result_sel := "001"; -- logical_result
|
|
|
|
elsif d_in.spr_info.ispmu = '1' then
|
|
|
|
v.e.result_sel := "100"; -- pmuspr_result
|
|
|
|
end if;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- See if any of the operands can get their value via the bypass path.
|
|
|
|
case gpr_a_bypass is
|
|
|
|
when "10" =>
|
|
|
|
v.e.read_data1 := execute_bypass.data;
|
|
|
|
when "11" =>
|
|
|
|
v.e.read_data1 := execute2_bypass.data;
|
|
|
|
when others =>
|
|
|
|
v.e.read_data1 := decoded_reg_a.data;
|
|
|
|
end case;
|
|
|
|
case gpr_b_bypass is
|
|
|
|
when "10" =>
|
|
|
|
v.e.read_data2 := execute_bypass.data;
|
|
|
|
when "11" =>
|
|
|
|
v.e.read_data2 := execute2_bypass.data;
|
|
|
|
when others =>
|
|
|
|
v.e.read_data2 := decoded_reg_b.data;
|
|
|
|
end case;
|
|
|
|
case gpr_c_bypass is
|
|
|
|
when "10" =>
|
|
|
|
v.e.read_data3 := execute_bypass.data;
|
|
|
|
when "11" =>
|
|
|
|
v.e.read_data3 := execute2_bypass.data;
|
|
|
|
when others =>
|
|
|
|
v.e.read_data3 := decoded_reg_c.data;
|
|
|
|
end case;
|
|
|
|
|
|
|
|
v.e.cr := c_in.read_cr_data;
|
|
|
|
if cr_bypass = "10" then
|
|
|
|
v.e.cr := execute_cr_bypass.data;
|
|
|
|
elsif cr_bypass = "11" then
|
|
|
|
v.e.cr := execute2_cr_bypass.data;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- issue control
|
|
|
|
control_valid_in <= d_in.valid;
|
|
|
|
control_sgl_pipe <= d_in.decode.sgl_pipe;
|
|
|
|
|
|
|
|
gpr_write_valid <= v.e.write_reg_enable;
|
|
|
|
gpr_write <= decoded_reg_o.reg;
|
|
|
|
|
|
|
|
gpr_a_read_valid <= decoded_reg_a.reg_valid;
|
|
|
|
gpr_a_read <= decoded_reg_a.reg;
|
|
|
|
|
|
|
|
gpr_b_read_valid <= decoded_reg_b.reg_valid;
|
|
|
|
gpr_b_read <= decoded_reg_b.reg;
|
|
|
|
|
|
|
|
gpr_c_read_valid <= decoded_reg_c.reg_valid;
|
|
|
|
gpr_c_read <= decoded_reg_c.reg;
|
|
|
|
|
|
|
|
cr_write_valid <= d_in.decode.output_cr or decode_rc(d_in.decode.rc, d_in.insn);
|
|
|
|
-- Since ops that write CR only write some of the fields,
|
|
|
|
-- any op that writes CR effectively also reads it.
|
|
|
|
cr_read_valid <= cr_write_valid or d_in.decode.input_cr;
|
|
|
|
|
|
|
|
v.e.valid := control_valid_out;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
if control_valid_out = '1' then
|
|
|
|
v.repeat := v.e.repeat and not r.repeat;
|
|
|
|
end if;
|
|
|
|
|
|
|
|
stall_out <= control_stall_out or v.repeat;
|
|
|
|
|
|
|
|
if rst = '1' or flush_in = '1' then
|
|
|
|
v.e := Decode2ToExecute1Init;
|
core: Implement quadword loads and stores
This implements the lq, stq, lqarx and stqcx. instructions.
These instructions all access two consecutive GPRs; for example the
"lq %r6,0(%r3)" instruction will load the doubleword at the address
in R3 into R7 and the doubleword at address R3 + 8 into R6. To cope
with having two GPR sources or destinations, the instruction gets
repeated at the decode2 stage, that is, for each lq/stq/lqarx/stqcx.
coming in from decode1, two instructions get sent out to execute1.
For these instructions, the RS or RT register gets modified on one
of the iterations by setting the LSB of the register number. In LE
mode, the first iteration uses RS|1 or RT|1 and the second iteration
uses RS or RT. In BE mode, this is done the other way around. In
order for decode2 to know what endianness is currently in use, we
pass the big_endian flag down from icache through decode1 to decode2.
This is always in sync with what execute1 is using because only rfid
or an interrupt can change MSR[LE], and those operations all cause
a flush and redirect.
There is now an extra column in the decode tables in decode1 to
indicate whether the instruction needs to be repeated. Decode1 also
enforces the rule that lq with RT = RT and lqarx with RA = RT or
RB = RT are illegal.
Decode2 now passes a 'repeat' flag and a 'second' flag to execute1,
and execute1 passes them on to loadstore1. The 'repeat' flag is set
for both iterations of a repeated instruction, and 'second' is set
on the second iteration. Execute1 does not take asynchronous or
trace interrupts on the second iteration of a repeated instruction.
Loadstore1 uses 'next_addr' for the second iteration of a repeated
load/store so that we access the second doubleword of the memory
operand. Thus loadstore1 accesses the doublewords in increasing
memory order. For 16-byte loads this means that the first iteration
writes GPR RT|1. It is possible that RA = RT|1 (this is a legal
but non-preferred form), meaning that if the memory operand was
misaligned, the first iteration would overwrite RA but then the
second iteration might take a page fault, leading to corrupted state.
To avoid that possibility, 16-byte loads in LE mode take an
alignment interrupt if the operand is not 16-byte aligned. (This
is the case anyway for lqarx, and we enforce it for lq as well.)
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
4 years ago
|
|
|
v.repeat := '0';
|
|
|
|
end if;
|
|
|
|
|
|
|
|
-- Update registers
|
|
|
|
rin <= v;
|
|
|
|
|
|
|
|
-- Update outputs
|
|
|
|
e_out <= r.e;
|
|
|
|
end process;
|
|
|
|
|
|
|
|
d2_log: if LOG_LENGTH > 0 generate
|
|
|
|
signal log_data : std_ulogic_vector(9 downto 0);
|
|
|
|
begin
|
|
|
|
dec2_log : process(clk)
|
|
|
|
begin
|
|
|
|
if rising_edge(clk) then
|
|
|
|
log_data <= r.e.nia(5 downto 2) &
|
|
|
|
r.e.valid &
|
|
|
|
stopped_out &
|
|
|
|
stall_out &
|
|
|
|
(gpr_a_bypass(1) or gpr_a_bypass(0)) &
|
|
|
|
(gpr_b_bypass(1) or gpr_b_bypass(0)) &
|
|
|
|
(gpr_c_bypass(1) or gpr_c_bypass(0));
|
|
|
|
end if;
|
|
|
|
end process;
|
|
|
|
log_out <= log_data;
|
|
|
|
end generate;
|
|
|
|
|
|
|
|
end architecture behaviour;
|