1 |
2 |
kdv |
/*
|
2 |
|
|
* fifo_size.v
|
3 |
|
|
*
|
4 |
|
|
* Copyright (c) 2007 Koen De Vleeschauwer.
|
5 |
|
|
*
|
6 |
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
|
7 |
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
8 |
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
9 |
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
|
10 |
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
11 |
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
12 |
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
13 |
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
14 |
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
15 |
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
16 |
|
|
* SUCH DAMAGE.
|
17 |
|
|
*/
|
18 |
|
|
|
19 |
|
|
/*
|
20 |
|
|
* fifo_size.v
|
21 |
|
|
*
|
22 |
|
|
* Dimensioning of various fifos. Tuning parameters; handle with care.
|
23 |
|
|
*
|
24 |
|
|
* Summary:
|
25 |
|
|
*
|
26 |
|
|
* motcomp throttles motcomp_addr_gen (via fifos_almost_full) when dst_wr_almost_full || fwd_wr_addr_almost_full || bwd_wr_addr_almost_full .
|
27 |
|
|
* This is tuneable via ADDR_THRESHOLD.
|
28 |
|
|
* ADDR_THRESHOLD has to be big enough to allow motcomp_addr_gen to generate all memory requests for a complete macroblock.
|
29 |
|
|
*
|
30 |
|
|
* framestore_request stops handling requests from fwd motion compensation when fwd_wr_dta_almost_full
|
31 |
|
|
* framestore_request stops handling requests from bwd motion compensation when bwd_wr_dta_almost_full
|
32 |
|
|
* framestore_request stops handling requests from display when disp_wr_dta_almost_full
|
33 |
|
|
* framestore_request stops handling requests for circular video buffer reads when vbr_wr_almost_full
|
34 |
|
|
* This is tuneable via DTA_THRESHOLD.
|
35 |
|
|
* If memory latency is high, DTA_THRESHOLD should be high as well.
|
36 |
|
|
*
|
37 |
|
|
* framestore_request stops handling requests when mem_req fills up (mem_req_wr_almost_full).
|
38 |
|
|
* This is tuneable via MEM_THRESHOLD.
|
39 |
|
|
*
|
40 |
|
|
* memory controller mem_ctl stops handling requests when memory results fifo almost full (mem_res_wr_almost_full).
|
41 |
|
|
* This is tuneable via MEM_THRESHOLD.
|
42 |
|
|
*
|
43 |
|
|
* framestore_response stops draining memory results fifo mem_res when fwd_wr_dta_full, bwd_wr_dta_full, disp_wr_dta_full or vbr_wr_full.
|
44 |
|
|
*
|
45 |
|
|
* mem_tag fifo size is smaller than mem_req or mem_res; hence the number of
|
46 |
|
|
* memory requests "in flight" - including memory latency - is at most mem_tag fifo size.
|
47 |
|
|
* This allows one to dimension DTA_THRESHOLD: if DTA_THRESHOLD is smaller
|
48 |
|
|
* than mem_tag fifo size, you run the risk of overflowing a data fifo.
|
49 |
|
|
* Indeed, framestore_request will stop issuing read requests when the data fifo has
|
50 |
|
|
* less than DTA_THRESHOLD space left; but since there may be up to mem_tag
|
51 |
|
|
* fifo size (= 2**MEMTAG_DEPTH) requests already queued for execution,
|
52 |
|
|
* the pending requests may overflow the data fifo if DTA_THRESHOLD < 2**MEMTAG_DEPTH.
|
53 |
|
|
* Hence always choose DTA_THRESHOLD > 2**MEMTAG_DEPTH.
|
54 |
|
|
*
|
55 |
|
|
* Remark: fifo sizes in this file are influenced by dual-port ram sizes available in FPGA's, typically 18 or 36 kbit.
|
56 |
|
|
* As such, fifo sizes in this file tend not to be minimal fifo sizes.
|
57 |
|
|
*/
|
58 |
|
|
|
59 |
|
|
|
60 |
|
|
/*
|
61 |
|
|
* dct_coeff fifo. 31 bits wide.
|
62 |
|
|
* Run/Length Values fifo from vld. Input for rld.
|
63 |
|
|
*/
|
64 |
|
|
|
65 |
|
|
parameter
|
66 |
|
|
RLD_DEPTH = 9'd7, // one 4:2:0 macroblock = 6 blocks at 64 run/length values per block maximum = 384 entries maximum
|
67 |
|
|
RLD_THRESHOLD = 9'd2,
|
68 |
|
|
|
69 |
|
|
/*
|
70 |
|
|
* predict_err_fifo. 72 bits wide.
|
71 |
|
|
* Inverse Discrete Cosine Transform Output. Contains prediction error.
|
72 |
|
|
*/
|
73 |
|
|
|
74 |
|
|
PREDICT_DEPTH = 9'd8,
|
75 |
|
|
PREDICT_THRESHOLD = 9'd64, // big enough so 1 macroblock ( 6 blocks @ 8 rows each ) fits.
|
76 |
|
|
|
77 |
|
|
/*
|
78 |
|
|
* mvec fifo. 206 bits wide.
|
79 |
|
|
* prediction motion vector fifo from vld. Input for motvec.
|
80 |
|
|
*/
|
81 |
|
|
|
82 |
|
|
MVEC_DEPTH = 9'd3,
|
83 |
|
|
MVEC_THRESHOLD = 9'd2,
|
84 |
|
|
|
85 |
|
|
/*
|
86 |
|
|
*
|
87 |
|
|
*/
|
88 |
|
|
|
89 |
|
|
ADDR_DEPTH = 9'd8,
|
90 |
|
|
ADDR_THRESHOLD = 9'd8,
|
91 |
|
|
DTA_DEPTH = 9'd8,
|
92 |
|
|
DTA_THRESHOLD = 9'd64,
|
93 |
|
|
MOTCOMP_ADDR_THRESHOLD = 9'd144, /* enough for motcomp_addrgen to produce all reads necessary to process a complete macroblock
|
94 |
|
|
number of addresses produced by motcomp_addrgen = no. of lumi blocks * lumi_rows * columns + no. of chromi blocks * max. chromi rows * colums = 4 * 9 * 2 + 2 * 10 * 2 = 112 (see motcomp_addrgen)
|
95 |
|
|
number of addresses in the mem_addr pipe: 13 (13 tages, numbered 0 to 12)
|
96 |
|
|
together: 112 + 13 = 125.
|
97 |
|
|
144: safety margin, just in case.
|
98 |
|
|
*/
|
99 |
|
|
|
100 |
|
|
/*
|
101 |
|
|
* Circular Video Buffer. vbuf_write_fifo and vbuf_read_fifo are both 64 bits wide.
|
102 |
|
|
* from stream input to getbits.
|
103 |
|
|
*/
|
104 |
|
|
|
105 |
|
|
VBUF_WR_DEPTH = DTA_DEPTH,
|
106 |
|
|
VBUF_WR_THRESHOLD = 9'd128,
|
107 |
|
|
VBUF_RD_DEPTH = DTA_DEPTH,
|
108 |
|
|
VBUF_RD_THRESHOLD = 9'd32,
|
109 |
|
|
|
110 |
|
|
/*
|
111 |
|
|
* fwd_reader. addr fifo is 22 bits wide; data fifo is 64 bits wide.
|
112 |
|
|
* Reads the data for forward motion compensation.
|
113 |
|
|
* Two fifo's: one sending addresses to be read to the frame store;
|
114 |
|
|
* one receiving data read from the frame store.
|
115 |
|
|
*/
|
116 |
|
|
|
117 |
|
|
FWD_ADDR_DEPTH = ADDR_DEPTH,
|
118 |
|
|
FWD_ADDR_THRESHOLD = MOTCOMP_ADDR_THRESHOLD,
|
119 |
|
|
FWD_DTA_DEPTH = DTA_DEPTH,
|
120 |
|
|
FWD_DTA_THRESHOLD = DTA_THRESHOLD, // less than half of dta fifo size
|
121 |
|
|
|
122 |
|
|
/*
|
123 |
|
|
* bwd_reader. addr fifo is 22 bits wide; data fifo is 64 bits wide.
|
124 |
|
|
* Reads the data for backward motion compensation.
|
125 |
|
|
* Two fifo's: one sending addresses to be read to the frame store;
|
126 |
|
|
* one receiving data read from the frame store.
|
127 |
|
|
*/
|
128 |
|
|
|
129 |
|
|
BWD_ADDR_DEPTH = ADDR_DEPTH,
|
130 |
|
|
BWD_ADDR_THRESHOLD = MOTCOMP_ADDR_THRESHOLD,
|
131 |
|
|
BWD_DTA_DEPTH = DTA_DEPTH,
|
132 |
|
|
BWD_DTA_THRESHOLD = DTA_THRESHOLD,
|
133 |
|
|
|
134 |
|
|
/*
|
135 |
|
|
* dst_fifo. 35 bits wide.
|
136 |
|
|
* Motion compensation. Queues the addresses where the reconstructed pixels need to be written
|
137 |
|
|
* until prediction error, forward and backward motion compensation data are available.
|
138 |
|
|
*/
|
139 |
|
|
|
140 |
|
|
DST_DEPTH = ADDR_DEPTH,
|
141 |
|
|
DST_THRESHOLD = MOTCOMP_ADDR_THRESHOLD,
|
142 |
|
|
|
143 |
|
|
/*
|
144 |
|
|
* recon_writer. 86 bits wide.
|
145 |
|
|
* Motion compensation. Writes reconstructed pixels to the frame store.
|
146 |
|
|
*/
|
147 |
|
|
|
148 |
|
|
RECON_DEPTH = DTA_DEPTH,
|
149 |
|
|
RECON_THRESHOLD = DTA_THRESHOLD,
|
150 |
|
|
|
151 |
|
|
/*
|
152 |
|
|
* disp_reader. addr fifo is 22 bits wide; data fifo is 64 bits wide.
|
153 |
|
|
* Reads pixels from the frame store for displaying.
|
154 |
|
|
* Two fifo's: one sending addresses to be read to the frame store;
|
155 |
|
|
* one receiving data read from the frame store.
|
156 |
|
|
*/
|
157 |
|
|
|
158 |
|
|
DISP_ADDR_DEPTH = ADDR_DEPTH,
|
159 |
|
|
DISP_ADDR_THRESHOLD= 9'd32, // about 8 times RESAMPLE_THRESHOLD
|
160 |
|
|
DISP_DTA_DEPTH = DTA_DEPTH, // disp_reader data fifo should never be empty
|
161 |
|
|
DISP_DTA_THRESHOLD = DTA_THRESHOLD, // less than half of dta fifo size.
|
162 |
|
|
|
163 |
|
|
/*
|
164 |
|
|
* resample_fifo. 3 bits wide.
|
165 |
|
|
* Chroma resampling. Fifo from resample_addr to resample_dta.
|
166 |
|
|
*/
|
167 |
|
|
|
168 |
|
|
RESAMPLE_DEPTH = 9'd8,
|
169 |
|
|
RESAMPLE_THRESHOLD = 9'd4,
|
170 |
|
|
|
171 |
|
|
/*
|
172 |
|
|
* pixel_fifo. 35 bits wide.
|
173 |
|
|
* From the decoding process to the display process
|
174 |
|
|
*/
|
175 |
|
|
|
176 |
|
|
PIXEL_DEPTH = 9'd10,
|
177 |
|
|
PIXEL_THRESHOLD = 9'd32,
|
178 |
|
|
|
179 |
|
|
/*
|
180 |
|
|
* osd_writer. 86 bits wide.
|
181 |
|
|
* On-Screen Display. Writes on-screen display to the frame store.
|
182 |
|
|
*/
|
183 |
|
|
|
184 |
|
|
OSD_DEPTH = 9'd5,
|
185 |
|
|
OSD_THRESHOLD = 9'd8,
|
186 |
|
|
|
187 |
|
|
/*
|
188 |
|
|
* threshold to make framestore_request stop writing before mem_request_fifo, mem_tag_fifo or mem_response_fifo overflow.
|
189 |
|
|
*/
|
190 |
|
|
|
191 |
|
|
MEM_THRESHOLD = 9'd16,
|
192 |
|
|
|
193 |
|
|
/*
|
194 |
|
|
* mem_request_fifo. 88 bits wide.
|
195 |
|
|
* Memory subsystem. Sends read, write and refresh commands to the memory controller.
|
196 |
|
|
*/
|
197 |
|
|
|
198 |
|
|
MEMREQ_DEPTH = 9'd6,
|
199 |
|
|
MEMREQ_THRESHOLD = MEM_THRESHOLD,
|
200 |
|
|
|
201 |
|
|
/*
|
202 |
|
|
* mem_tag_fifo. 3 bits wide.
|
203 |
|
|
* Memory subsystem. Queues tags of read commands sent to the memory controller.
|
204 |
|
|
*/
|
205 |
|
|
|
206 |
|
|
MEMTAG_DEPTH = 9'd5,
|
207 |
|
|
MEMTAG_THRESHOLD = MEM_THRESHOLD,
|
208 |
|
|
|
209 |
|
|
/*
|
210 |
|
|
* mem_response_fifo. 64 bits wide.
|
211 |
|
|
* Memory subsystem. Receives data read from the memory controller.
|
212 |
|
|
*/
|
213 |
|
|
|
214 |
|
|
MEMRESP_DEPTH = 9'd7,
|
215 |
|
|
MEMRESP_THRESHOLD = 9'd64;
|
216 |
|
|
|
217 |
|
|
/* not truncated */
|