cores
/
microwatt
mirror of https://github.com/antonblanchard/microwatt
3
1
Fork 1
Code Issues Projects Releases Wiki Activity

loadstore1: Make r1.req.addr not depend on l_in.valid

Browse Source 
Some critical path reports showed r1.req.addr depending on l_in.valid,
which then depended ultimately on the dcache's r1.ls_valid.  In fact
we can update r1.req.addr (and other fields of r1.req, except for
r1.req.valid) independently of l_in.valid as long as busy = 0.
We do also need to preserve r1.req.addr0 when l_in.valid = 0, so we
pull it out of r1.req and store it separately in r1.

Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
pull/327/head
Paul Mackerras 4 years ago
parent 2224b28c2c
commit bb5f356386
1 changed files with 42 additions and 37 deletions
Show Stats Download Patch File Download Diff File

79
loadstore1.vhdl
Unescape Escape View File

@ -68,7 +68,6 @@ architecture behave of loadstore1 is
noop : std_ulogic; noop : std_ulogic;
mode_32bit : std_ulogic; mode_32bit : std_ulogic;
addr : std_ulogic_vector(63 downto 0); addr : std_ulogic_vector(63 downto 0);
addr0 : std_ulogic_vector(63 downto 0);
byte_sel : std_ulogic_vector(7 downto 0); byte_sel : std_ulogic_vector(7 downto 0);
second_bytes : std_ulogic_vector(7 downto 0); second_bytes : std_ulogic_vector(7 downto 0);
store_data : std_ulogic_vector(63 downto 0); store_data : std_ulogic_vector(63 downto 0);
@ -99,7 +98,7 @@ architecture behave of loadstore1 is
constant request_init : request_t := (valid => '0', dc_req => '0', load => '0', store => '0', tlbie => '0', constant request_init : request_t := (valid => '0', dc_req => '0', load => '0', store => '0', tlbie => '0',
dcbz => '0', read_spr => '0', write_spr => '0', mmu_op => '0', dcbz => '0', read_spr => '0', write_spr => '0', mmu_op => '0',
instr_fault => '0', load_zero => '0', do_update => '0', noop => '0', instr_fault => '0', load_zero => '0', do_update => '0', noop => '0',
mode_32bit => '0', addr => (others => '0'), addr0 => (others => '0'), mode_32bit => '0', addr => (others => '0'),
byte_sel => x"00", second_bytes => x"00", byte_sel => x"00", second_bytes => x"00",
store_data => (others => '0'), instr_tag => instr_tag_init, store_data => (others => '0'), instr_tag => instr_tag_init,
write_reg => 7x"00", length => x"0", write_reg => 7x"00", length => x"0",
@ -115,6 +114,7 @@ architecture behave of loadstore1 is
type reg_stage1_t is record type reg_stage1_t is record
req : request_t; req : request_t;
issued : std_ulogic; issued : std_ulogic;
addr0 : std_ulogic_vector(63 downto 0);
end record; end record;


type reg_stage2_t is record type reg_stage2_t is record
@ -125,6 +125,7 @@ architecture behave of loadstore1 is
wait_mmu : std_ulogic; wait_mmu : std_ulogic;
one_cycle : std_ulogic; one_cycle : std_ulogic;
wr_sel : std_ulogic_vector(1 downto 0); wr_sel : std_ulogic_vector(1 downto 0);
addr0 : std_ulogic_vector(63 downto 0);
end record; end record;


type reg_stage3_t is record type reg_stage3_t is record
@ -361,7 +362,7 @@ begin


-- Translate a load/store instruction into the internal request format -- Translate a load/store instruction into the internal request format
-- XXX this should only depend on l_in, but actually depends on -- XXX this should only depend on l_in, but actually depends on
-- r1.req.addr0 as well (in the l_in.second = 1 case). -- r1.addr0 as well (in the l_in.second = 1 case).
loadstore1_in: process(all) loadstore1_in: process(all)
variable v : request_t; variable v : request_t;
variable lsu_sum : std_ulogic_vector(63 downto 0); variable lsu_sum : std_ulogic_vector(63 downto 0);
@ -402,23 +403,21 @@ begin
end if; end if;


addr := lsu_sum; addr := lsu_sum;

if l_in.second = '1' then if l_in.second = '1' then
if l_in.update = '0' then if l_in.update = '0' then
-- for the second half of a 16-byte transfer, -- for the second half of a 16-byte transfer,
-- use the previous address plus 8. -- use the previous address plus 8.
addr := std_ulogic_vector(unsigned(r1.req.addr0(63 downto 3)) + 1) & r1.req.addr0(2 downto 0); addr := std_ulogic_vector(unsigned(r1.addr0(63 downto 3)) + 1) & r1.addr0(2 downto 0);
else else
-- for an update-form load, use the previous address -- for an update-form load, use the previous address
-- as the value to write back to RA. -- as the value to write back to RA.
addr := r1.req.addr0; addr := r1.addr0;
end if; end if;
end if; end if;
if l_in.mode_32bit = '1' then if l_in.mode_32bit = '1' then
addr(63 downto 32) := (others => '0'); addr(63 downto 32) := (others => '0');
end if; end if;
v.addr := addr; v.addr := addr;
v.addr0 := addr;


-- XXX Temporary hack. Mark the op as non-cachable if the address -- XXX Temporary hack. Mark the op as non-cachable if the address
-- is the form 0xc------- for a real-mode access. -- is the form 0xc------- for a real-mode access.
@ -516,28 +515,31 @@ begin
variable v : reg_stage1_t; variable v : reg_stage1_t;
variable req : request_t; variable req : request_t;
variable dcreq : std_ulogic; variable dcreq : std_ulogic;
variable addr : std_ulogic_vector(63 downto 0); variable issue : std_ulogic;
begin begin
v := r1; v := r1;
dcreq := '0'; issue := '0';
req := req_in;
if flushing = '1' then if busy = '0' then
-- Make this a no-op request rather than simply invalid. req := req_in;
-- It will never get to stage 3 since there is a request ahead of v.issued := '0';
-- it with align_intr = 1. if flushing = '1' then
req.dc_req := '0'; -- Make this a no-op request rather than simply invalid.
-- It will never get to stage 3 since there is a request ahead of
-- it with align_intr = 1.
req.dc_req := '0';
end if;
issue := l_in.valid and req.dc_req;
if l_in.valid = '1' then
v.addr0 := req.addr;
end if;
else
req := r1.req;
end if; end if;


-- Note that l_in.valid is gated with busy inside execute1 if r1.req.valid = '1' then
if l_in.valid = '1' then
dcreq := req.dc_req and stage1_issue_enable and not d_in.error and not dc_stall;
v.req := req;
v.issued := dcreq;
elsif r1.req.valid = '1' then
if r1.req.dc_req = '1' and r1.issued = '0' then if r1.req.dc_req = '1' and r1.issued = '0' then
req := r1.req; issue := '1';
dcreq := stage1_issue_enable and not dc_stall and not d_in.error;
v.issued := dcreq;
elsif r1.issued = '1' and d_in.error = '1' then elsif r1.issued = '1' and d_in.error = '1' then
v.issued := '0'; v.issued := '0';
elsif stage2_busy_next = '0' then elsif stage2_busy_next = '0' then
@ -545,23 +547,25 @@ begin
-- in r1 will go into r2 -- in r1 will go into r2
if r1.req.dc_req = '1' and r1.req.two_dwords = '1' and r1.req.dword_index = '0' then if r1.req.dc_req = '1' and r1.req.two_dwords = '1' and r1.req.dword_index = '0' then
-- construct the second request for a misaligned access -- construct the second request for a misaligned access
v.req.dword_index := '1'; req.dword_index := '1';
v.req.addr := std_ulogic_vector(unsigned(r1.req.addr(63 downto 3)) + 1) & "000"; req.addr := std_ulogic_vector(unsigned(r1.req.addr(63 downto 3)) + 1) & "000";
if r1.req.mode_32bit = '1' then if r1.req.mode_32bit = '1' then
v.req.addr(32) := '0'; req.addr(32) := '0';
end if; end if;
v.req.byte_sel := r1.req.second_bytes; req.byte_sel := r1.req.second_bytes;
v.issued := stage1_issue_enable and not dc_stall; issue := '1';
dcreq := stage1_issue_enable and not dc_stall;
req := v.req;
else
v.req.valid := '0';
end if; end if;
end if; end if;
end if; end if;
if r3in.interrupt = '1' then if r3in.interrupt = '1' then
v.req.valid := '0'; req.valid := '0';
dcreq := '0'; issue := '0';
end if;

v.req := req;
dcreq := issue and stage1_issue_enable and not d_in.error and not dc_stall;
if issue = '1' then
v.issued := dcreq;
end if; end if;


stage1_req <= req; stage1_req <= req;
@ -584,7 +588,7 @@ begin


-- Byte reversing and rotating for stores. -- Byte reversing and rotating for stores.
-- Done in the second cycle (the cycle after l_in.valid = 1). -- Done in the second cycle (the cycle after l_in.valid = 1).
byte_offset := unsigned(r1.req.addr0(2 downto 0)); byte_offset := unsigned(r1.addr0(2 downto 0));
for i in 0 to 7 loop for i in 0 to 7 loop
k := (to_unsigned(i, 3) - byte_offset) xor r1.req.brev_mask; k := (to_unsigned(i, 3) - byte_offset) xor r1.req.brev_mask;
j := to_integer(k) * 8; j := to_integer(k) * 8;
@ -594,6 +598,7 @@ begin
if stage3_busy_next = '0' and if stage3_busy_next = '0' and
(r1.req.valid = '0' or r1.issued = '1' or r1.req.dc_req = '0') then (r1.req.valid = '0' or r1.issued = '1' or r1.req.dc_req = '0') then
v.req := r1.req; v.req := r1.req;
v.addr0 := r1.addr0;
v.req.store_data := store_data; v.req.store_data := store_data;
v.wait_dc := r1.req.valid and r1.req.dc_req and not r1.req.load_sp and v.wait_dc := r1.req.valid and r1.req.dc_req and not r1.req.load_sp and
not (r1.req.two_dwords and not r1.req.dword_index); not (r1.req.two_dwords and not r1.req.dword_index);
@ -881,7 +886,7 @@ begin
write_data := sprval; write_data := sprval;
when "01" => when "01" =>
-- update reg -- update reg
write_data := r2.req.addr0; write_data := r2.addr0;
when "10" => when "10" =>
-- lfs result -- lfs result
write_data := load_dp_data; write_data := load_dp_data;

Write Preview
Loading…
Cancel
Save