1 |
26 |
zuofu |
--ECE395 GPU:
|
2 |
|
|
--Blitter Subunit
|
3 |
|
|
--=====================================================
|
4 |
|
|
--Designed by:
|
5 |
|
|
--Zuofu Cheng
|
6 |
|
|
--James Cavanaugh
|
7 |
|
|
--Eric Sands
|
8 |
|
|
--
|
9 |
|
|
--of the University of Illinois at Urbana Champaign
|
10 |
|
|
--under the direction of Dr. Lippold Haken
|
11 |
|
|
--====================================================
|
12 |
|
|
--
|
13 |
|
|
--Heavily based off of HDL examples provided by XESS Corporation
|
14 |
|
|
--www.xess.com
|
15 |
|
|
--
|
16 |
|
|
--Based in part on Doug Hodson's work which in turn
|
17 |
|
|
--was based off of the XSOC from Gray Research LLC.
|
18 |
|
|
--
|
19 |
|
|
--
|
20 |
|
|
--release under the GNU General Public License
|
21 |
|
|
--and kindly hosted by www.opencores.org
|
22 |
|
|
|
23 |
|
|
library IEEE, UNISIM;
|
24 |
|
|
use IEEE.std_logic_1164.all;
|
25 |
|
|
use IEEE.numeric_std.all;
|
26 |
|
|
use UNISIM.VComponents.all;
|
27 |
|
|
use WORK.common.all;
|
28 |
|
|
use WORK.xsasdram.all;
|
29 |
|
|
use WORK.sdram.all;
|
30 |
|
|
use WORK.vga_pckg.all;
|
31 |
|
|
use WORK.fifo_cc_pckg.all;
|
32 |
|
|
use IEEE.STD_LOGIC_ARITH.ALL;
|
33 |
|
|
use IEEE.STD_LOGIC_UNSIGNED.ALL;
|
34 |
|
|
|
35 |
|
|
|
36 |
|
|
package Blitter_pckg is
|
37 |
|
|
component Blitter
|
38 |
|
|
generic(
|
39 |
|
|
FREQ : natural := 50_000; -- operating frequency in KHz
|
40 |
|
|
PIPE_EN : boolean := true; -- enable pipelined operations
|
41 |
|
|
DATA_WIDTH : natural := 16; -- host & SDRAM data width
|
42 |
|
|
ADDR_WIDTH : natural := 23 -- host-side address width
|
43 |
|
|
);
|
44 |
|
|
port(
|
45 |
|
|
clk : in std_logic; -- master clock
|
46 |
|
|
rst : in std_logic; -- reset for this entity
|
47 |
|
|
rd : out std_logic; -- initiate read operation
|
48 |
|
|
wr : out std_logic; -- initiate write operation
|
49 |
|
|
opBegun : in std_logic; -- read/write/self-refresh op begun (clocked)
|
50 |
|
|
earlyopBegun : in std_logic;
|
51 |
|
|
done : in std_logic; -- read or write operation is done
|
52 |
|
|
rddone : in std_logic; -- read operation is done
|
53 |
|
|
rdPending : in std_logic; -- read operation is not done
|
54 |
|
|
Addr : out std_logic_vector(ADDR_WIDTH-1 downto 0); -- address to SDRAM
|
55 |
|
|
DIn : out std_logic_vector(DATA_WIDTH-1 downto 0); -- data to dualport to SDRAM
|
56 |
|
|
DOut : in std_logic_vector(DATA_WIDTH-1 downto 0); -- data from dualport to SDRAM
|
57 |
|
|
|
58 |
|
|
blit_begin : in std_logic;
|
59 |
|
|
source_address : in std_logic_vector(ADDR_WIDTH-1 downto 0);
|
60 |
|
|
source_lines : in std_logic_vector(15 downto 0);
|
61 |
|
|
|
62 |
|
|
target_address : in std_logic_vector(ADDR_WIDTH-1 downto 0);
|
63 |
|
|
line_size : in std_logic_vector(11 downto 0);
|
64 |
|
|
alphaOp : in std_logic; -- if true then current operation is a blend alphaOp
|
65 |
|
|
front_buffer : in std_logic; -- if false, then we are writing to back buffer
|
66 |
|
|
blit_done : out std_logic --line has been completely copied
|
67 |
|
|
);
|
68 |
|
|
end component Blitter;
|
69 |
|
|
end package Blitter_pckg;
|
70 |
|
|
|
71 |
|
|
library IEEE, UNISIM;
|
72 |
|
|
use IEEE.std_logic_1164.all;
|
73 |
|
|
use IEEE.numeric_std.all;
|
74 |
|
|
use UNISIM.VComponents.all;
|
75 |
|
|
use WORK.common.all;
|
76 |
|
|
use WORK.xsasdram.all;
|
77 |
|
|
use WORK.sdram.all;
|
78 |
|
|
use WORK.vga_pckg.all;
|
79 |
|
|
use WORK.fifo_cc_pckg.all;
|
80 |
|
|
use IEEE.STD_LOGIC_ARITH.ALL;
|
81 |
|
|
use IEEE.STD_LOGIC_UNSIGNED.ALL;
|
82 |
|
|
|
83 |
|
|
---- Uncomment the following library declaration if instantiating
|
84 |
|
|
---- any Xilinx primitives in this code.
|
85 |
|
|
--library UNISIM;
|
86 |
|
|
--use UNISIM.VComponents.all;
|
87 |
|
|
|
88 |
|
|
entity Blitter is
|
89 |
|
|
generic(
|
90 |
|
|
FREQ : natural := 50_000; -- operating frequency in KHz
|
91 |
|
|
PIPE_EN : boolean := true; -- enable pipelined operations
|
92 |
|
|
DATA_WIDTH : natural := 16; -- host & SDRAM data width
|
93 |
|
|
ADDR_WIDTH : natural := 23 -- host-side address width
|
94 |
|
|
);
|
95 |
|
|
port(
|
96 |
|
|
clk : in std_logic; -- master clock
|
97 |
|
|
rst : in std_logic; -- reset for this entity
|
98 |
|
|
rd : out std_logic; -- initiate read operation
|
99 |
|
|
wr : out std_logic; -- initiate write operation
|
100 |
|
|
opBegun : in std_logic; -- read/write/self-refresh op begun (clocked)
|
101 |
|
|
earlyopBegun : in std_logic;
|
102 |
|
|
done : in std_logic; -- read or write operation is done
|
103 |
|
|
rddone : in std_logic; -- read operation is done
|
104 |
|
|
rdPending : in std_logic; -- read operation is not done
|
105 |
|
|
Addr : out std_logic_vector(ADDR_WIDTH-1 downto 0); -- address to SDRAM
|
106 |
|
|
DIn : out std_logic_vector(DATA_WIDTH-1 downto 0); -- data to dualport to SDRAM
|
107 |
|
|
DOut : in std_logic_vector(DATA_WIDTH-1 downto 0); -- data from dualport to SDRAM
|
108 |
|
|
|
109 |
|
|
blit_begin : in std_logic;
|
110 |
|
|
source_address : in std_logic_vector(ADDR_WIDTH-1 downto 0);
|
111 |
|
|
source_lines : in std_logic_vector(15 downto 0);
|
112 |
|
|
|
113 |
|
|
target_address : in std_logic_vector(ADDR_WIDTH-1 downto 0);
|
114 |
|
|
line_size : in std_logic_vector(11 downto 0);
|
115 |
|
|
alphaOp : in std_logic; -- if true then current operation is a blend alphaOp
|
116 |
|
|
front_buffer : in std_logic; -- if false, then we are writing to back buffer
|
117 |
|
|
blit_done : out std_logic --line has been completely copied
|
118 |
|
|
);
|
119 |
|
|
end Blitter;
|
120 |
|
|
|
121 |
|
|
architecture Behavioral of Blitter is
|
122 |
|
|
|
123 |
|
|
-- states of the memory tester state machine
|
124 |
|
|
type blitState is (
|
125 |
|
|
STANDBY,
|
126 |
|
|
INIT1, -- init
|
127 |
|
|
INIT2,
|
128 |
|
|
READ, -- load queue with line
|
129 |
|
|
EMPTY_PIPE, -- empty read pipeline
|
130 |
|
|
READ_B, -- read contents of VRAM for blend
|
131 |
|
|
EMPTY_PIPE_B, -- empty read pipeline again
|
132 |
|
|
WRITE1, -- compare memory contents with pseudo-random data
|
133 |
|
|
WRITE2,
|
134 |
|
|
STOP -- stop and indicate memory status
|
135 |
|
|
);
|
136 |
|
|
signal state_r, state_x : blitState; -- state register and next state
|
137 |
|
|
|
138 |
|
|
-- registers
|
139 |
|
|
signal addr_r, addr_x : unsigned(addr'range); -- address register
|
140 |
|
|
|
141 |
|
|
signal s_addr_r, s_addr_x : std_logic_vector(ADDR_WIDTH - 1 downto 0); --per line start address
|
142 |
|
|
signal t_addr_r, t_addr_x : std_logic_vector(ADDR_WIDTH - 1 downto 0); --per line target address
|
143 |
|
|
|
144 |
|
|
signal current_line_r, current_line_x : std_logic_vector(15 downto 0);
|
145 |
|
|
signal count_r, count_x : std_logic_vector(11 downto 0);
|
146 |
|
|
signal tot_pix_r, tot_pix_x : std_logic_vector(11 downto 0);
|
147 |
|
|
signal blend_data_r, blend_data_x : std_logic_vector(15 downto 0);
|
148 |
|
|
|
149 |
|
|
-- internal signals
|
150 |
|
|
signal rd_q : std_logic;
|
151 |
|
|
signal wr_q : std_logic;
|
152 |
|
|
signal empty_q : std_logic;
|
153 |
|
|
signal reset_q : std_logic;
|
154 |
|
|
signal out_q : std_logic_vector(15 downto 0);
|
155 |
|
|
signal level_q: std_logic_vector(7 downto 0);
|
156 |
|
|
|
157 |
|
|
signal rd_b : std_logic;
|
158 |
|
|
signal wr_b : std_logic;
|
159 |
|
|
signal empty_b : std_logic;
|
160 |
|
|
signal reset_b : std_logic;
|
161 |
|
|
signal out_b : std_logic_vector(15 downto 0);
|
162 |
|
|
signal level_b: std_logic_vector(7 downto 0);
|
163 |
|
|
|
164 |
|
|
signal q_write : std_logic;
|
165 |
|
|
signal b_write : std_logic;
|
166 |
|
|
|
167 |
|
|
begin
|
168 |
|
|
|
169 |
|
|
-- fifo buffer
|
170 |
|
|
u1 : fifo_cc
|
171 |
|
|
port map(
|
172 |
|
|
clk=>clk,
|
173 |
|
|
rst=>reset_q,
|
174 |
|
|
rd=>rd_q,
|
175 |
|
|
wr=>wr_q,
|
176 |
|
|
data_in=>dOut,
|
177 |
|
|
data_out=>out_q,
|
178 |
|
|
full=>open,
|
179 |
|
|
empty=>empty_q,
|
180 |
|
|
level=>level_q
|
181 |
|
|
);
|
182 |
|
|
|
183 |
|
|
u2 : fifo_cc
|
184 |
|
|
port map(
|
185 |
|
|
clk=>clk,
|
186 |
|
|
rst=>reset_q, --hack
|
187 |
|
|
rd=>rd_b,
|
188 |
|
|
wr=>wr_b,
|
189 |
|
|
data_in=>dOut,
|
190 |
|
|
data_out=>out_b,
|
191 |
|
|
full=>open,
|
192 |
|
|
empty=>empty_b,
|
193 |
|
|
level=>level_b
|
194 |
|
|
);
|
195 |
|
|
|
196 |
|
|
-- connect internal registers to external busses
|
197 |
|
|
|
198 |
|
|
-- memory address bus driven by memory address register
|
199 |
|
|
addr <= std_logic_vector(addr_r); -- linear memory addressing
|
200 |
|
|
wr_q <= rdDone and q_write; -- whenever something is causing rdDone to go high, we need to queue it
|
201 |
|
|
wr_b <= rdDone and b_write;
|
202 |
|
|
|
203 |
|
|
--if not doing an AlphaOp, we can directly copy pixel, wire vram input to blend result register
|
204 |
|
|
dIn <= blend_data_r when alphaOp = YES else out_q;
|
205 |
|
|
|
206 |
|
|
-- memory test controller state machine operations
|
207 |
|
|
combinatorial : process(state_r, addr_r, current_line_r,
|
208 |
|
|
s_addr_r, t_addr_r,
|
209 |
|
|
earlyOpbegun, rdPending,
|
210 |
|
|
source_address, target_address,
|
211 |
|
|
empty_q, source_lines, line_size, count_r,
|
212 |
|
|
alphaOp, level_q, level_b, tot_pix_r, blit_begin, front_buffer) -- *********** added by Eric
|
213 |
|
|
begin
|
214 |
|
|
|
215 |
|
|
-- default operations (do nothing unless explicitly stated in the following case statement)
|
216 |
|
|
rd <= NO; -- no memory write
|
217 |
|
|
wr <= NO; -- no memory read
|
218 |
|
|
rd_q <= NO;
|
219 |
|
|
rd_b <= NO;
|
220 |
|
|
reset_q <= NO;
|
221 |
|
|
blit_done <= NO;
|
222 |
|
|
q_write <= NO;
|
223 |
|
|
b_write <= NO;
|
224 |
|
|
|
225 |
|
|
current_line_x <= current_line_r;
|
226 |
|
|
s_addr_x <= s_addr_r;
|
227 |
|
|
t_addr_x <= t_addr_r;
|
228 |
|
|
count_x <= count_r;
|
229 |
|
|
tot_pix_x <= tot_pix_r;
|
230 |
|
|
addr_x <= addr_r; -- next address is the same as current address
|
231 |
|
|
state_x <= state_r;
|
232 |
|
|
blend_data_x <= blend_data_r;
|
233 |
|
|
|
234 |
|
|
-- **** compute the next state and operations ****
|
235 |
|
|
case state_r is
|
236 |
|
|
|
237 |
|
|
------------------------------------------------------
|
238 |
|
|
-- initialize blitter
|
239 |
|
|
------------------------------------------------------
|
240 |
|
|
when STANDBY => -- goto this state at beginning of operation
|
241 |
|
|
if (blit_begin = YES) then
|
242 |
|
|
state_x <= INIT1;
|
243 |
|
|
end if;
|
244 |
|
|
|
245 |
|
|
when INIT1 => -- latch in blit parameters
|
246 |
|
|
if (front_buffer = YES) then
|
247 |
|
|
t_addr_x <= target_address;
|
248 |
|
|
else
|
249 |
|
|
t_addr_x <= target_address + x"009600";
|
250 |
|
|
end if;
|
251 |
|
|
|
252 |
|
|
s_addr_x <= source_address;
|
253 |
|
|
tot_pix_x <= line_size;
|
254 |
|
|
current_line_x <= x"0000";
|
255 |
|
|
count_x <= x"000";
|
256 |
|
|
reset_q <= YES;
|
257 |
|
|
blit_done <= NO;
|
258 |
|
|
state_x <= INIT2;
|
259 |
|
|
|
260 |
|
|
when INIT2 => -- this state happens for every line
|
261 |
|
|
addr_x <= unsigned(s_addr_r);
|
262 |
|
|
count_x <= x"000";
|
263 |
|
|
q_write <= YES;
|
264 |
|
|
state_x <= READ;
|
265 |
|
|
|
266 |
|
|
when READ => -- Read data in from the source address into the FIFO
|
267 |
|
|
q_write <= YES;
|
268 |
|
|
rd <= YES; -- actual write to FIFO is straight through assignment
|
269 |
|
|
if (earlyOpBegun = YES) then -- which is somewhere above
|
270 |
|
|
addr_x <= addr_r + 1; -- increment address read next memory location
|
271 |
|
|
count_x <= count_r + 1;
|
272 |
|
|
end if;
|
273 |
|
|
if count_r = tot_pix_r then
|
274 |
|
|
state_x <= EMPTY_PIPE;
|
275 |
|
|
end if;
|
276 |
|
|
|
277 |
|
|
when EMPTY_PIPE =>
|
278 |
|
|
q_write <= YES;
|
279 |
|
|
if (rdPending = NO) then
|
280 |
|
|
count_x <= x"000";
|
281 |
|
|
addr_x <= unsigned(t_addr_r); -- set address to target
|
282 |
|
|
if (alphaOp = NO) then -- if not doing alpha operation then we can write the pixels
|
283 |
|
|
state_x <= WRITE1;
|
284 |
|
|
else
|
285 |
|
|
state_x <= READ_B; -- otherwise we need to read the current contents of the VRAM
|
286 |
|
|
end if;
|
287 |
|
|
end if;
|
288 |
|
|
|
289 |
|
|
when READ_B =>
|
290 |
|
|
b_write <= YES;
|
291 |
|
|
rd <= YES; -- read from VRAM into buffer FIFO_B
|
292 |
|
|
if (earlyOpBegun = YES) then
|
293 |
|
|
addr_x <= addr_r + 1;
|
294 |
|
|
count_x <= count_r + 1;
|
295 |
|
|
end if;
|
296 |
|
|
if (count_r = tot_pix_r) then
|
297 |
|
|
state_x <= EMPTY_PIPE_B;
|
298 |
|
|
end if;
|
299 |
|
|
|
300 |
|
|
when EMPTY_PIPE_B =>
|
301 |
|
|
b_write <= YES;
|
302 |
|
|
if (rdPending = NO) then
|
303 |
|
|
count_x <= x"000";
|
304 |
|
|
addr_x <= unsigned(t_addr_r); -- set address to target
|
305 |
|
|
rd_q <= YES;
|
306 |
|
|
rd_b <= YES;
|
307 |
|
|
state_x <= WRITE1;
|
308 |
|
|
|
309 |
|
|
end if;
|
310 |
|
|
|
311 |
|
|
when WRITE1 =>
|
312 |
|
|
count_x <= x"000";
|
313 |
|
|
rd_q <= YES;
|
314 |
|
|
rd_b <= YES;
|
315 |
|
|
--perform blending here (synchronous)
|
316 |
|
|
if (out_q(15 downto 8) = x"E7") then
|
317 |
|
|
blend_data_x(15 downto 8) <= out_b(15 downto 8);
|
318 |
|
|
else
|
319 |
|
|
blend_data_x(15 downto 8 )<= out_q(15 downto 8);
|
320 |
|
|
end if;
|
321 |
|
|
|
322 |
|
|
if (out_q(7 downto 0) = x"E7") then
|
323 |
|
|
blend_data_x(7 downto 0) <= out_b(7 downto 0);
|
324 |
|
|
else
|
325 |
|
|
blend_data_x(7 downto 0)<= out_q(7 downto 0);
|
326 |
|
|
end if;
|
327 |
|
|
|
328 |
|
|
state_x <= WRITE2;
|
329 |
|
|
|
330 |
|
|
when WRITE2 => -- write to memory
|
331 |
|
|
wr <= YES; -- with data from FIFO
|
332 |
|
|
|
333 |
|
|
if (earlyOpBegun = YES) then
|
334 |
|
|
rd_q <= YES;
|
335 |
|
|
rd_b <= YES;
|
336 |
|
|
|
337 |
|
|
--perform blending here (synchronous)
|
338 |
|
|
if (out_q(15 downto 8) = x"E7") then
|
339 |
|
|
blend_data_x(15 downto 8) <= out_b(15 downto 8);
|
340 |
|
|
else
|
341 |
|
|
blend_data_x(15 downto 8 )<= out_q(15 downto 8);
|
342 |
|
|
end if;
|
343 |
|
|
|
344 |
|
|
if (out_q(7 downto 0) = x"E7") then
|
345 |
|
|
blend_data_x(7 downto 0) <= out_b(7 downto 0);
|
346 |
|
|
else
|
347 |
|
|
blend_data_x(7 downto 0)<= out_q(7 downto 0);
|
348 |
|
|
end if;
|
349 |
|
|
|
350 |
|
|
if empty_q /= YES then -- if we still have more data to write to VRAM
|
351 |
|
|
addr_x <= addr_r + 1; -- increment address
|
352 |
|
|
count_x <= count_r + 1;
|
353 |
|
|
else
|
354 |
|
|
current_line_x <= current_line_r + 1; -- we're done with a whole line
|
355 |
|
|
state_x <= STOP;
|
356 |
|
|
end if;
|
357 |
|
|
end if;
|
358 |
|
|
|
359 |
|
|
when others => -- stop and unreachable states
|
360 |
|
|
if (current_line_r = source_lines) then
|
361 |
|
|
blit_done <= YES; -- Operation is complete, so we just sit here
|
362 |
|
|
else -- all per line stuff neeeds to be reset here
|
363 |
|
|
count_x <= x"000";
|
364 |
|
|
s_addr_x <= s_addr_r + x"0000A0";
|
365 |
|
|
t_addr_x <= t_addr_r + x"0000A0";
|
366 |
|
|
reset_q <= YES;
|
367 |
|
|
state_x <= INIT2;
|
368 |
|
|
end if;
|
369 |
|
|
end case;
|
370 |
|
|
|
371 |
|
|
end process;
|
372 |
|
|
|
373 |
|
|
|
374 |
|
|
-- update the registers of the memory tester controller
|
375 |
|
|
update : process(clk)
|
376 |
|
|
begin
|
377 |
|
|
if (clk'event and clk = '1') then
|
378 |
|
|
if rst = YES then
|
379 |
|
|
state_r <= STANDBY;
|
380 |
|
|
else
|
381 |
|
|
addr_r <= addr_x;
|
382 |
|
|
state_r <= state_x;
|
383 |
|
|
current_line_r <= current_line_x;
|
384 |
|
|
s_addr_r <= s_addr_x;
|
385 |
|
|
t_addr_r <= t_addr_x;
|
386 |
|
|
count_r <= count_x;
|
387 |
|
|
tot_pix_r <= tot_pix_x;
|
388 |
|
|
blend_data_r <= blend_data_x;
|
389 |
|
|
end if;
|
390 |
|
|
end if;
|
391 |
|
|
end process;
|
392 |
|
|
|
393 |
|
|
end Behavioral;
|