1 The Acorn Electron ULA
2 ======================
3
4 Principal Design and Feature Constraints
5 ----------------------------------------
6
7 The features of the ULA are limited in sophistication by the amount of time
8 and resources that can be allocated to each activity supporting the
9 fundamental features and obligations of the unit. Maintaining a screen display
10 based on the contents of RAM itself requires the ULA to have exclusive access
11 to various hardware resources for a significant period of time.
12
13 Whilst other elements of the ULA can in principle run in parallel with the
14 display refresh activity, they cannot also access the RAM at the same time.
15 Consequently, other features that might use the RAM must accept a reduced
16 allocation of that resource in comparison to a hypothetical architecture where
17 concurrent RAM access is possible at all times.
18
19 Thus, the principal constraint for many features is bandwidth. The duration of
20 access to hardware resources is one aspect of this; the rate at which such
21 resources can be accessed is another. For example, the RAM is not fast enough
22 to support access more frequently than one byte per 2MHz cycle, and for screen
23 modes involving 80 bytes of screen data per scanline, there are no free cycles
24 for anything other than the production of pixel output during the active
25 scanline periods.
26
27 Another constraint is imposed by the method of RAM access provided by the ULA.
28 The ULA is able to access RAM by fetching 4 bits at a time and thus managing
29 to transfer 8 bits within a single 2MHz cycle, this being sufficient to
30 provide display data for the most demanding screen modes. However, this
31 mechanism's timing requirements are beyond the capabilities of the CPU when
32 running at 2MHz.
33
34 Consequently, the CPU will only ever be able to access RAM via the ULA at
35 1MHz, even when the ULA is not accessing the RAM. Fortunately, when needing to
36 refresh the display, the ULA is still able to make use of the idle part of
37 each 1MHz cycle (or, rather, the idle 2MHz cycle unused by the CPU) to itself
38 access the RAM at a rate of 1 byte per 1MHz cycle (or 1 byte every other 2MHz
39 cycle), thus supporting the less demanding screen modes.
40
41 Timing
42 ------
43
44 According to 15.3.2 in the Advanced User Guide, there are 312 scanlines, 256
45 of which are used to generate pixel data. At 50Hz, this means that 128 cycles
46 are spent on each scanline (2000000 cycles / 50 = 40000 cycles; 40000 cycles /
47 312 ~= 128 cycles). This is consistent with the observation that each scanline
48 requires at most 80 bytes of data, and that the ULA is apparently busy for 40
49 out of 64 microseconds in each scanline.
50
51 (In fact, since the ULA is seeking to provide an image for an interlaced
52 625-line display, there are in fact two "fields" involved, one providing 312
53 scanlines and one providing 313 scanlines. See below for a description of the
54 video system.)
55
56 Access to RAM involves accessing four 64Kb dynamic RAM devices (IC4 to IC7,
57 each providing two bits of each byte) using two cycles within the 500ns period
58 of the 2MHz clock to complete each access operation. Since the CPU and ULA
59 have to take turns in accessing the RAM in MODE 4, 5 and 6, the CPU must
60 effectively run at 1MHz (since every other 500ns period involves the ULA
61 accessing RAM) during transfers of screen data.
62
63 The CPU is driven by an external clock (IC8) whose 16MHz frequency is divided
64 by the ULA (IC1) depending on the screen mode in use. Each 16MHz cycle is
65 approximately 62.5ns. To access the memory, the following patterns
66 corresponding to 16MHz cycles are required:
67
68 Time (ns): 0-------------- 500------------- ...
69 2 MHz cycle: 0 1 ...
70 16 MHz cycle: 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7 ...
71 /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\ ...
72 ~RAS: /---\___________/---\___________ ...
73 ~CAS: /-----\___/-\___/-----\___/-\___ ...
74 Address events: A B C A B C ...
75 Data events: F S F S ...
76
77 ~RAS ops: 1 0 1 0 ...
78 ~CAS ops: 1 0 1 0 1 0 1 0 ...
79
80 Address ops: a b c a b c ...
81 Data ops: s f s f ...
82
83 ~WE: ......W ...
84 PHI OUT: \_______________/--------------- ...
85 CPU (RAM): L D ...
86 RnW: R ...
87
88 PHI OUT: \_______/-------\_______/------- ...
89 CPU (ROM): L D L D ...
90 RnW: R R ...
91
92 ~RAS must be high for 100ns, ~CAS must be high for 50ns.
93 ~RAS must be low for 150ns, ~CAS must be low for 90ns.
94 Data is available 150ns after ~RAS goes low, 90ns after ~CAS goes low.
95
96 Here, "A" and "B" respectively indicate the row and first column addresses
97 being latched into the RAM (on a negative edge for ~RAS and ~CAS
98 respectively), and "C" indicates the second column address being latched into
99 the RAM. Presumably, the first and second half-bytes can be read at "F" and
100 "S" respectively, and the row and column addresses must be made available at
101 "a" and "b" (and "c") respectively at the latest. Data can be read at "f" and
102 "s" for the first and second half-bytes respectively.
103
104 For the CPU, "L" indicates the point at which an address is taken from the CPU
105 address bus, on a negative edge of PHI OUT, with "D" being the point at which
106 data may either be read or be asserted for writing, on a positive edge of PHI
107 OUT. Here, PHI OUT is driven at 1MHz. Given that ~WE needs to be driven low
108 for writing or high for reading, and thus propagates RnW from the CPU, this
109 would need to be done before data would be retrieved and, according to the
110 TM4164EC4 datasheet, even as late as the column address is presented and ~CAS
111 brought low.
112
113 The TM4164EC4-15 has a row address access time of 150ns (maximum) and a column
114 address access time of 90ns (maximum), which appears to mean that ~RAS must be
115 held low for at least 150ns and that ~CAS must be held low for at least 90ns
116 before data becomes available. 150ns is 2.4 cycles (at 16MHz) and 90ns is 1.44
117 cycles. Thus, "A" to "F" is 2.5 cycles, "B" to "F" is 1.5 cycles, "C" to "S"
118 is 1.5 cycles.
119
120 Note that the Service Manual refers to the negative edge of RAS and CAS, but
121 the datasheet for the similar TM4164EC4 product shows latching on the negative
122 edge of ~RAS and ~CAS. It is possible that the Service Manual also intended to
123 communicate the latter behaviour. In the TM4164EC4 datasheet, it appears that
124 "page mode" provides the appropriate behaviour for that particular product.
125
126 The CPU, when accessing the RAM alone, apparently does not make use of the
127 vacated "slot" that the ULA would otherwise use (when interleaving accesses in
128 MODE 4, 5 and 6). It only employs a full 2MHz access frequency to memory when
129 accessing ROM (and potentially sideways RAM). The principal limitation is the
130 amount of time needed between issuing an address and receiving an entire byte
131 from the RAM, which is approximately 7 cycles (at 16MHz): much longer than the
132 4 cycles that would be required for 2MHz operation.
133
134 See: Acorn Electron Advanced User Guide
135 See: Acorn Electron Service Manual
136 http://chrisacorns.computinghistory.org.uk/docs/Acorn/Manuals/Acorn_ElectronSM.pdf
137 See: http://mdfs.net/Docs/Comp/Electron/Techinfo.htm
138 See: http://stardot.org.uk/forums/viewtopic.php?p=120438#p120438
139
140 CPU Clock Notes
141 ---------------
142
143 "The 6502 receives an external square-wave clock input signal on pin 37, which
144 is usually labeled PHI0. [...] This clock input is processed within the 6502
145 to form two clock outputs: PHI1 and PHI2 (pins 3 and 39, respectively). PHI2
146 is essentially a copy of PHI0; more specifically, PHI2 is PHI0 after it's been
147 through two inverters and a push-pull amplifier. The same network of
148 transistors within the 6502 which generates PHI2 is also tied to PHI1, and
149 generates PHI1 as the inverse of PHI0. The reason why PHI1 and PHI2 are made
150 available to external devices is so that they know when they can access the
151 CPU. When PHI1 is high, this means that external devices can read from the
152 address bus or data bus; when PHI2 is high, this means that external devices
153 can write to the data bus."
154
155 See: http://lateblt.livejournal.com/88105.html
156
157 "The 6502 has a synchronous memory bus where the master clock is divided into
158 two phases (Phase 1 and Phase 2). The address is always generated during Phase
159 1 and all memory accesses take place during Phase 2."
160
161 See: http://www.jmargolin.com/vgens/vgens.htm
162
163 Thus, the inverse of PHI OUT provides the "other phase" of the clock. "During
164 Phase 1" means when PHI0 - really PHI2 - is high and "during Phase 2" means
165 when PHI1 is high.
166
167 Bandwidth Figures
168 -----------------
169
170 Using an observation of 128 2MHz cycles per scanline, 256 active lines and 312
171 total lines, with 80 cycles occurring in the active periods of display
172 scanlines, the following bandwidth calculations can be performed:
173
174 Total theoretical maximum:
175 128 cycles * 312 lines
176 = 39936 bytes
177
178 MODE 0, 1, 2:
179 ULA: 80 cycles * 256 lines
180 = 20480 bytes
181 CPU: 48 cycles / 2 * 256 lines
182 + 128 cycles / 2 * (312 - 256) lines
183 = 9728 bytes
184
185 MODE 3:
186 ULA: 80 cycles * 24 rows * 8 lines
187 = 15360 bytes
188 CPU: 48 cycles / 2 * 24 rows * 8 lines
189 + 128 cycles / 2 * (312 - (24 rows * 8 lines))
190 = 12288 bytes
191
192 MODE 4, 5:
193 ULA: 40 cycles * 256 lines
194 = 10240 bytes
195 CPU: (40 cycles + 48 cycles / 2) * 256 lines
196 + 128 cycles / 2 * (312 - 256) lines
197 = 19968 bytes
198
199 MODE 6:
200 ULA: 40 cycles * 24 rows * 8 lines
201 = 7680 bytes
202 CPU: (40 cycles + 48 cycles / 2) * 24 rows * 8 lines
203 + 128 cycles / 2 * (312 - (24 rows * 8 lines))
204 = 19968 bytes
205
206 Here, the division of 2 for CPU accesses is performed to indicate that the CPU
207 only uses every other access opportunity even in uncontended periods. See the
208 2MHz RAM Access enhancement below for bandwidth calculations that consider
209 this limitation removed.
210
211 Video Timing
212 ------------
213
214 According to 8.7 in the Service Manual, and the PAL Wikipedia page,
215 approximately 4.7µs is used for the sync pulse, 5.7µs for the "back porch"
216 (including the "colour burst"), and 1.65µs for the "front porch", totalling
217 12.05µs and thus leaving 51.95µs for the active video signal for each
218 scanline. As the Service Manual suggests in the oscilloscope traces, the
219 display information is transmitted more or less centred within the active
220 video period since the ULA will only be providing pixel data for 40µs in each
221 scanline.
222
223 Each 62.5ns cycle happens to correspond to 64µs divided by 1024, meaning that
224 each scanline can be divided into 1024 cycles, although only 640 at most are
225 actively used to provide pixel data. Pixel data production should only occur
226 within a certain period on each scanline, approximately 262 cycles after the
227 start of hsync:
228
229 active video period = 51.95µs
230 pixel data period = 40µs
231 total silent period = 51.95µs - 40µs = 11.95µs
232 silent periods (before and after) = 11.95µs / 2 = 5.975µs
233 hsync and back porch period = 4.7µs + 5.7µs = 10.4µs
234 time before pixel data period = 10.4µs + 5.975µs = 16.375µs
235 pixel data period start cycle = 16.375µs / 62.5ns = 262
236
237 By choosing a number divisible by 8, the RAM access mechanism can be
238 synchronised with the pixel production. Thus, 256 is a more appropriate start
239 cycle, where the HS (horizontal sync) signal corresponding to the 4µs sync
240 pulse (or "normal sync" pulse as described by the "PAL TV timing and voltages"
241 document) occurs at cycle 0.
242
243 To summarise:
244
245 HS signal starts at cycle 0 on each horizontal scanline
246 HS signal ends approximately 4µs later at cycle 64
247 Pixel data starts approximately 12µs later at cycle 256
248
249 "Re: Electron Memory Contention" provides measurements that appear consistent
250 with these calculations.
251
252 The "vertical blanking period", meaning the period before picture information
253 in each field is 25 lines out of 312 (or 313) and thus lasts for 1.6ms. Of
254 this, 2.5 lines occur before the vsync (field sync) which also lasts for 2.5
255 lines. Thus, the first visible scanline on the first field of a frame occurs
256 half way through the 23rd scanline period measured from the start of vsync
257 (indicated by "V" in the diagrams below):
258
259 10 20 23
260 Line in frame: 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8
261 Line from 1: 0 22 3
262 Line on screen: .:::::VVVVV::::: 12233445566
263 |_________________________________________________|
264 25 line vertical blanking period
265
266 In the second field of a frame, the first visible scanline coincides with the
267 24th scanline period measured from the start of line 313 in the frame:
268
269 310 336
270 Line in frame: 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
271 Line from 313: 0 23 4
272 Line on screen: 88:::::VVVVV:::: 11223344
273 288 | |
274 |_________________________________________________|
275 25 line vertical blanking period
276
277 In order to consider only full lines, we might consider the start of each
278 frame to occur 23 lines after the start of vsync.
279
280 Again, it is likely that pixel data production should only occur on scanlines
281 within a certain period on each frame. The "625/50" document indicates that
282 only a certain region is "safe" to use, suggesting a vertically centred region
283 with approximately 15 blank lines above and below the picture. However, the
284 "PAL TV timing and voltages" document suggests 28 blank lines above and below
285 the picture. This would centre the 256 lines within the 312 lines of each
286 field and thus provide a start of picture approximately 5.5 or 5 lines after
287 the end of the blanking period or 28 or 27.5 lines after the start of vsync.
288
289 To summarise:
290
291 CSYNC signal starts at cycle 0
292 CSYNC signal ends approximately 160µs (2.5 lines) later at cycle 2560
293 Start of line occurs approximately 1632µs (5.5 lines) later at cycle 28672
294
295 See: http://en.wikipedia.org/wiki/PAL
296 See: http://en.wikipedia.org/wiki/Analog_television#Structure_of_a_video_signal
297 See: The 625/50 PAL Video Signal and TV Compatible Graphics Modes
298 http://lipas.uwasa.fi/~f76998/video/modes/
299 See: PAL TV timing and voltages
300 http://www.retroleum.co.uk/electronics-articles/pal-tv-timing-and-voltages/
301 See: Line Standards
302 http://www.pembers.freeserve.co.uk/World-TV-Standards/Line-Standards.html
303 See: Horizontal Blanking Interval of 405-, 525-, 625- and 819-Line Standards
304 http://www.pembers.freeserve.co.uk/World-TV-Standards/HBI.pdf
305 See: Re: Electron Memory Contention
306 http://www.stardot.org.uk/forums/viewtopic.php?p=134109#p134109
307
308 RAM Integrated Circuits
309 -----------------------
310
311 Unicorn Electronics appears to offer 4164 RAM chips (as well as 6502 series
312 CPUs such as the 6502, 6502A, 6502B and 65C02). These 4164 devices are
313 available in 100ns (4164-100), 120ns (4164-120) and 150ns (4164-150) variants,
314 have 16 pins and address 65536 bits through a 1-bit wide channel. Similarly,
315 ByteDelight.com sell 4164 devices primarily for the ZX Spectrum.
316
317 The documentation for the Electron mentions 4164-15 RAM chips for IC4-7, and
318 the Samsung-produced KM41464 series is apparently equivalent to the Texas
319 Instruments 4164 chips presumably used in the Electron.
320
321 The TM4164EC4 series combines 4 64K x 1b units into a single package and
322 appears similar to the TM4164EA4 featured on the Electron's circuit diagram
323 (in the Advanced User Guide but not the Service Manual), and it also has 22
324 pins providing 3 additional inputs and 3 additional outputs over the 16 pins
325 of the individual 4164-15 modules, presumably allowing concurrent access to
326 the packaged memory units.
327
328 As far as currently available replacements are concerned, the NTE4164 is a
329 potential candidate: according to the Vetco Electronics entry, it is
330 supposedly a replacement for the TMS4164-15 amongst many other parts. Similar
331 parts include the NTE2164 and the NTE6664, both of which appear to have
332 largely the same performance and connection characteristics. Meanwhile, the
333 NTE21256 appears to be a 16-pin replacement with four times the capacity that
334 maintains the single data input and output pins. Using the NTE21256 as a
335 replacement for all ICs combined would be difficult because of the single bit
336 output.
337
338 Another device equivalent to the 4164-15 appears to be available under the
339 code 41662 from Jameco Electronics as the Siemens HYB 4164-2. The Jameco Web
340 site lists data sheets for other devices on the same page, but these are
341 different and actually appear to be provided under the 41574 product code (but
342 are listed under 41464-10) and appear to be replacements for the TM4164EC4:
343 the Samsung KM41464A-15 and NEC µPD41464 employ 18 pins, eliminating 4 pins by
344 employing 4 pins for both input and output.
345
346 Pins I/O pins Row access Column access
347 ---- -------- ---------- -------------
348 TM4164EC4 22 4 + 4 150ns (15) 90ns (15)
349 KM41464AP 18 4 150ns (15) 75ns (15)
350 NTE21256 16 1 + 1 150ns 75ns
351 HYB 4164-2 16 1 + 1 150ns 100ns
352 µPD41464 18 4 120ns (12) 60ns (12)
353
354 See: TM4164EC4 65,536 by 4-Bit Dynamic RAM Module
355 http://www.datasheetarchive.com/dl/Datasheets-112/DSAP0051030.pdf
356 See: Dynamic RAMS
357 http://www.unicornelectronics.com/IC/DYNAMIC.html
358 See: New old stock 8x 4164 chips
359 http://www.bytedelight.com/?product=8x-4164-chips-new-old-stock
360 See: KM4164B 64K x 1 Bit Dynamic RAM with Page Mode
361 http://images.ihscontent.net/vipimages/VipMasterIC/IC/SAMS/SAMSD020/SAMSD020-45.pdf
362 See: NTE2164 Integrated Circuit 65,536 X 1 Bit Dynamic Random Access Memory
363 http://www.vetco.net/catalog/product_info.php?products_id=2806
364 See: NTE4164 - IC-NMOS 64K DRAM 150NS
365 http://www.vetco.net/catalog/product_info.php?products_id=3680
366 See: NTE21256 - IC-256K DRAM 150NS
367 http://www.vetco.net/catalog/product_info.php?products_id=2799
368 See: NTE21256 262,144-Bit Dynamic Random Access Memory (DRAM)
369 http://www.nteinc.com/specs/21000to21999/pdf/nte21256.pdf
370 See: NTE6664 - IC-MOS 64K DRAM 150NS
371 http://www.vetco.net/catalog/product_info.php?products_id=5213
372 See: NTE6664 Integrated Circuit 64K-Bit Dynamic RAM
373 http://www.nteinc.com/specs/6600to6699/pdf/nte6664.pdf
374 See: 4164-150: MAJOR BRANDS
375 http://www.jameco.com/webapp/wcs/stores/servlet/Product_10001_10001_41662_-1
376 See: HYB 4164-1, HYB 4164-2, HYB 4164-3 65,536-Bit Dynamic Random Access Memory (RAM)
377 http://www.jameco.com/Jameco/Products/ProdDS/41662SIEMENS.pdf
378 See: KM41464A NMOS DRAM 64K x 4 Bit Dynamic RAM with Page Mode
379 http://www.jameco.com/Jameco/Products/ProdDS/41662SAM.pdf
380 See: NEC µ41464 65,536 x 4-Bit Dynamic NMOS RAM
381 http://www.jameco.com/Jameco/Products/ProdDS/41662NEC.pdf
382 See: 41464-10: MAJOR BRANDS
383 http://www.jameco.com/webapp/wcs/stores/servlet/Product_10001_10001_41574_-1
384
385 Interrupts
386 ----------
387
388 The ULA generates IRQs (maskable interrupts) according to certain conditions
389 and these conditions are controlled by location &FE00:
390
391 * Vertical sync (bottom of displayed screen)
392 * 50MHz real time clock
393 * Transmit data empty
394 * Receive data full
395 * High tone detect
396
397 The ULA is also used to clear interrupt conditions through location &FE05. Of
398 particular significance is bit 7, which must be set if an NMI (non-maskable
399 interrupt) has occurred and has thus suspended ULA access to memory, restoring
400 the normal function of the ULA.
401
402 ROM Paging
403 ----------
404
405 Accessing different ROMs involves bits 0 to 3 of &FE05. Some special ROM
406 mappings exist:
407
408 8 keyboard
409 9 keyboard (duplicate)
410 10 BASIC ROM
411 11 BASIC ROM (duplicate)
412
413 Paging in a ROM involves the following procedure:
414
415 1. Assert ROM page enable (bit 3) together with a ROM number n in bits 0 to
416 2, corresponding to ROM number 8+n, such that one of ROMs 12 to 15 is
417 selected.
418 2. Where a ROM numbered from 0 to 7 is to be selected, set bit 3 to zero
419 whilst writing the desired ROM number n in bits 0 to 2.
420
421 See: http://stardot.org.uk/forums/viewtopic.php?p=136686#p136686
422
423 Shadow/Expanded Memory
424 ----------------------
425
426 The Electron exposes all sixteen address lines and all eight data lines
427 through the expansion bus. Using such lines, it is possible to provide
428 additional memory - typically sideways ROM and RAM - on expansion cards and
429 through cartridges, although the official cartridge specification provides
430 fewer address lines and only seeks to provide access to memory in 16K units.
431
432 Various modifications and upgrades were developed to offer "turbo"
433 capabilities to the Electron, permitting the CPU to access a separate 8K of
434 RAM at 2MHz, presumably preventing access to the low 8K of RAM accessible via
435 the ULA through additional logic. However, an enhanced ULA might support
436 independent CPU access to memory over the expansion bus by allowing itself to
437 be discharged from providing access to memory, potentially for a range of
438 addresses, and for the CPU to communicate with external memory uninterrupted.
439
440 Sideways RAM/ROM and Upper Memory Access
441 ----------------------------------------
442
443 Although the ULA controls the CPU clock, effectively slowing or stopping the
444 CPU when the ULA needs to access screen memory, it is apparently able to allow
445 the CPU to access addresses of &8000 and above - the upper region of memory -
446 at 2MHz independently of any access to RAM that the ULA might be performing,
447 only blocking the CPU if it attempts to access addresses of &7FFF and below
448 during any ULA memory access - the lower region of memory - by stopping or
449 stalling its clock.
450
451 Thus, the ULA remains aware of the level of the A15 line, only inhibiting the
452 CPU clock if the line goes low, when the CPU is attempting to access the lower
453 region of memory.
454
455 Hardware Scrolling (and Enhancement)
456 ------------------------------------
457
458 On the standard ULA, &FE02 and &FE03 map to a 9 significant bits address with
459 the least significant 5 bits being zero, thus limiting the scrolling
460 resolution to 64 bytes. An enhanced ULA could support a resolution of 2 bytes
461 using the same layout of these addresses.
462
463 |--&FE02--------------| |--&FE03--------------|
464 XX XX 14 13 12 11 10 09 08 07 06 XX XX XX XX XX
465
466 XX 14 13 12 11 10 09 08 07 06 05 04 03 02 01 XX
467
468 Arguably, a resolution of 8 bytes is more useful, since the mapping of screen
469 memory to pixel locations is character oriented. A change in 8 bytes would
470 permit a horizontal scrolling resolution of 2 pixels in MODE 2, 4 pixels in
471 MODE 1 and 5, and 8 pixels in MODE 0, 3 and 6. This resolution is actually
472 observed on the BBC Micro (see 18.11.2 in the BBC Microcomputer Advanced User
473 Guide).
474
475 One argument for a 2 byte resolution is smooth vertical scrolling. A pitfall
476 of changing the screen address by 2 bytes is the change in the number of lines
477 from the initial and final character rows that need reading by the ULA, which
478 would need to maintain this state information (although this is a relatively
479 trivial change). Another pitfall is the complication that might be introduced
480 to software writing bitmaps of character height to the screen.
481
482 See: http://pastraiser.com/computers/acornelectron/acornelectron.html
483
484 Enhancement: Mode Layouts
485 -------------------------
486
487 Merely changing the screen memory mappings in order to have Archimedes-style
488 row-oriented screen addresses (instead of character-oriented addresses) could
489 be done for the existing modes, but this might not be sufficiently beneficial,
490 especially since accessing regions of the screen would involve incrementing
491 pointers by amounts that are inconvenient on an 8-bit CPU.
492
493 However, instead of using a Archimedes-style mapping, column-oriented screen
494 addresses could be more feasibly employed: incrementing the address would
495 reference the vertical screen location below the currently-referenced location
496 (just as occurs within characters using the existing ULA); instead of
497 returning to the top of the character row and referencing the next horizontal
498 location after eight bytes, the address would reference the next character row
499 and continue to reference locations downwards over the height of the screen
500 until reaching the bottom; at the bottom, the next location would be the next
501 horizontal location at the top of the screen.
502
503 In other words, the memory layout for the screen would resemble the following
504 (for MODE 2):
505
506 &3000 &3100 ... &7F00
507 &3001 &3101
508 ... ...
509 &3007
510 &3008
511 ...
512 ... ...
513 &30FF ... &7FFF
514
515 Since there are 256 pixel rows, each column of locations would be addressable
516 using the low byte of the address. Meanwhile, the high byte would be
517 incremented to address different columns. Thus, addressing screen locations
518 would become a lot more convenient and potentially much more efficient for
519 certain kinds of graphical output.
520
521 One potential complication with this simplified addressing scheme arises with
522 hardware scrolling. Vertical hardware scrolling by one pixel row (not supported
523 with the existing ULA) would be achieved by incrementing or decrementing the
524 screen start address; by one character row, it would involve adding or
525 subtracting 8. However, the ULA only supports multiples of 64 when changing the
526 screen start address. Thus, if such a scheme were to be adopted, three
527 additional bits would need to be supported in the screen start register (see
528 "Hardware Scrolling (and Enhancement)" for more details). However, horizontal
529 scrolling would be much improved even under the severe constraints of the
530 existing ULA: only adjustments of 256 to the screen start address would be
531 required to produce single-location scrolling of as few as two pixels in MODE 2
532 (four pixels in MODEs 1 and 5, eight pixels otherwise).
533
534 More disruptive is the effect of this alternative layout on software.
535 Presumably, compatibility with the BBC Micro was the primary goal of the
536 Electron's hardware design. With the character-oriented screen layout in
537 place, system software (and application software accessing the screen
538 directly) would be relying on this layout to run on the Electron with little
539 or no modification. Although it might have been possible to change the system
540 software to use this column-oriented layout instead, this would have incurred
541 a development cost and caused additional work porting things like games to the
542 Electron. Moreover, a separate branch of the software from that supporting the
543 BBC Micro and closer derivatives would then have needed maintaining.
544
545 The decision to use the character-oriented layout in the BBC Micro may have
546 been related to the choice of circuitry and to facilitate a convenient
547 hardware implementation, and by the time the Electron was planned, it was too
548 late to do anything about this somewhat unfortunate choice.
549
550 Pixel Layouts
551 -------------
552
553 The pixel layouts are as follows:
554
555 Modes Depth (bpp) Pixels (from bits)
556 ----- ----------- ------------------
557 0, 3, 4, 6 1 7 6 5 4 3 2 1 0
558 1, 5 2 73 62 51 40
559 2 4 7531 6420
560
561 Since the ULA reads a half-byte at a time, one might expect it to attempt to
562 produce pixels for every half-byte, as opposed to handling entire bytes.
563 However, the pixel layout is not conducive to producing pixels as soon as a
564 half-byte has been read for a given full-byte location: in 1bpp modes the
565 first four pixels can indeed be produced, but in 2bpp and 4bpp modes the pixel
566 data is spread across the entire byte in different ways.
567
568 An alternative arrangement might be as follows:
569
570 Modes Depth (bpp) Pixels (from bits)
571 ----- ----------- ------------------
572 0, 3, 4, 6 1 7 6 5 4 3 2 1 0
573 1, 5 2 76 54 32 10
574 2 4 7654 3210
575
576 Just as the mode layouts were presumably decided by compatibility with the BBC
577 Micro, the pixel layouts will have been maintained for similar reasons.
578 Unfortunately, this layout prevents any optimisation of the ULA for handling
579 half-byte pixel data generally.
580
581 Enhancement: The Missing MODE 4
582 -------------------------------
583
584 The Electron inherits its screen mode selection from the BBC Micro, where MODE
585 3 is a text version of MODE 0, and where MODE 6 is a text version of MODE 4.
586 Neither MODE 3 nor MODE 6 is a genuine character-based text mode like MODE 7,
587 however, and they are merely implemented by skipping two scanlines in every
588 ten after the eight required to produce a character line. Thus, such modes
589 provide a 24-row display.
590
591 In principle, nothing prevents this "text mode" effect being applied to other
592 modes. The 20-column modes are not well-suited to displaying text, which
593 leaves MODE 1 which, unlike MODEs 3 and 6, can display 4 colours rather than
594 2. Although the need for a non-monochrome 40-column text mode is addressed by
595 MODE 7 on the BBC Micro, the Electron lacks such a mode.
596
597 If the 4-colour, 24-row variant of MODE 1 were to be provided, logically it
598 would occupy MODE 4 instead of the current MODE 4:
599
600 Screen mode Size (kilobytes) Colours Rows Resolution
601 ----------- ---------------- ------- ---- ----------
602 0 20 2 32 640x256
603 1 20 4 32 320x256
604 2 20 16 32 160x256
605 3 16 2 24 640x256
606 4 (new) 16 4 24 320x256
607 4 (old) 10 2 32 320x256
608 5 10 4 32 160x256
609 6 8 2 24 320x256
610
611 Thus, for increasing mode numbers, the size of each mode would be the same or
612 less than the preceding mode.
613
614 Enhancement: 2MHz RAM Access
615 ----------------------------
616
617 Given that the CPU and ULA both access RAM at 2MHz, but given that the CPU
618 when not competing with the ULA only accesses RAM every other 2MHz cycle (as
619 if the ULA still needed to access the RAM), one useful enhancement would be a
620 mechanism to let the CPU take over the ULA cycles outside the ULA's period of
621 activity comparable to the way the ULA takes over the CPU cycles in MODE 0 to
622 3.
623
624 Thus, the RAM access cycles would resemble the following in MODE 0 to 3:
625
626 Upon a transition from display cycles: UUUUCCCC (instead of UUUUC_C_)
627 On a non-display line: CCCCCCCC (instead of C_C_C_C_)
628
629 In MODE 4 to 6:
630
631 Upon a transition from display cycles: CUCUCCCC (instead of CUCUC_C_)
632 On a non-display line: CCCCCCCC (instead of C_C_C_C_)
633
634 This would improve CPU bandwidth as follows:
635
636 Standard ULA Enhanced ULA
637 MODE 0, 1, 2 9728 bytes 19456 bytes
638 MODE 3 12288 bytes 24576 bytes
639 MODE 4, 5 19968 bytes 29696 bytes
640 MODE 6 19968 bytes 32256 bytes
641
642 (Here, the uncontended 2MHz bandwidth for a display period would be 39936
643 bytes, being 128 cycles per line over 312 lines.)
644
645 With such an enhancement, MODE 0 to 3 experience a doubling of CPU bandwidth
646 because all access opportunities to RAM are doubled. Meanwhile, in the other
647 modes, some CPU accesses occur alongside ULA accesses and thus cannot be
648 doubled, but the CPU bandwidth increase is still significant.
649
650 Unfortunately, the mechanism for accessing the RAM is too slow to provide data
651 within the time constraints of 2MHz operation. There is no time remaining in a
652 2MHz cycle for the CPU to receive and process any retrieved data.
653
654 Enhancement: Region Blanking
655 ----------------------------
656
657 The problem of permitting character-oriented blitting in programs whilst
658 scrolling the screen by sub-character amounts could be mitigated by permitting
659 a region of the display to be blank, such as the final lines of the display.
660 Consider the following vertical scrolling by 2 bytes that would cause an
661 initial character row of 6 lines and a final character row of 2 lines:
662
663 6 lines - initial, partial character row
664 248 lines - 31 complete rows
665 2 lines - final, partial character row
666
667 If a routine were in use that wrote 8 line bitmaps to the partial character
668 row now split in two, it would be advisable to hide one of the regions in
669 order to prevent content appearing in the wrong place on screen (such as
670 content meant to appear at the top "leaking" onto the bottom). Blanking 6
671 lines would be sufficient, as can be seen from the following cases.
672
673 Scrolling up by 2 lines:
674
675 6 lines - initial, partial character row
676 240 lines - 30 complete rows
677 4 lines - part of 1 complete row
678 -----------------------------------------------------------------
679 4 lines - part of 1 complete row (hidden to maintain 250 lines)
680 2 lines - final, partial character row (hidden)
681
682 Scrolling down by 2 lines:
683
684 2 lines - initial, partial character row
685 248 lines - 31 complete rows
686 ----------------------------------------------------------
687 6 lines - final, partial character row (hidden)
688
689 Thus, in this case, region blanking would impose a 250 line display with the
690 bottom 6 lines blank.
691
692 See the description of the display suspend enhancement for a more efficient
693 way of blanking lines than merely blanking the palette whilst allowing the CPU
694 to perform useful work during the blanking period.
695
696 To control the blanking or suspending of lines at the top and bottom of the
697 display, a memory location could be dedicated to the task: the upper 4 bits
698 could define a blanking region of up to 16 lines at the top of the screen,
699 whereas the lower 4 bits could define such a region at the bottom of the
700 screen. If more lines were required, two locations could be employed, allowing
701 the top and bottom regions to occupy the entire screen.
702
703 Enhancement: Screen Height Adjustment
704 -------------------------------------
705
706 The height of the screen could be configurable in order to reduce screen
707 memory consumption. This is not quite done in MODE 3 and 6 since the start of
708 the screen appears to be rounded down to the nearest page, but by reducing the
709 height by amounts more than a page, savings would be possible. For example:
710
711 Screen width Depth Height Bytes per line Saving in bytes Start address
712 ------------ ----- ------ -------------- --------------- -------------
713 640 1 252 80 320 &3140 -> &3100
714 640 1 248 80 640 &3280 -> &3200
715 320 1 240 40 640 &5A80 -> &5A00
716 320 2 240 80 1280 &3500
717
718 Screen Mode Selection
719 ---------------------
720
721 Bits 3, 4 and 5 of address &FE*7 control the selected screen mode. For a wider
722 range of modes, the other bits of &FE*7 (related to sound, cassette
723 input/output and the Caps Lock LED) would need to be reassigned and bit 0
724 potentially being made available for use.
725
726 Enhancement: Palette Definition
727 -------------------------------
728
729 Since all memory accesses go via the ULA, an enhanced ULA could employ more
730 specific addresses than &FE*X to perform enhanced functions. For example, the
731 palette control is done using &FE*8-F and merely involves selecting predefined
732 colours, whereas an enhanced ULA could support the redefinition of all 16
733 colours using specific ranges such as &FE18-F (colours 0 to 7) and &FE28-F
734 (colours 8 to 15), where a single byte might provide 8 bits per pixel colour
735 specifications similar to those used on the Archimedes.
736
737 The principal limitation here is actually the hardware: the Electron has only
738 a single output line for each of the red, green and blue channels, and if
739 those outputs are strictly digital and can only be set to a "high" and "low"
740 value, then only the existing eight colours are possible. If a modern ULA were
741 able to output analogue values (or values at well-defined points between the
742 high and low values, such as the half-on value supported by the Amstrad CPC
743 series), it would still need to be assessed whether the circuitry could
744 successfully handle and propagate such values. Various sources indicate that
745 only "TTL levels" are supported by the RGB output circuit, and since there are
746 74LS08 AND logic gates involved in the RGB component outputs from the ULA, it
747 is likely that the ULA is expected to provide only "high" or "low" values.
748
749 Short of adding extra outputs from the ULA (either additional red, green and
750 blue outputs or a combined intensity output), another approach might involve
751 some kind of modulation where an output value might be encoded in multiple
752 pulses at a higher frequency than the pixel frequency. However, this would
753 demand additional circuitry outside the ULA, and component RGB monitors would
754 probably not be able to take advantage of this feature; only UHF and composite
755 video devices (the latter with the composite video colour support enabled on
756 the Electron's circuit board) would potentially benefit.
757
758 Flashing Colours
759 ----------------
760
761 According to the Advanced User Guide, "The cursor and flashing colours are
762 entirely generated in software: This means that all of the logical to physical
763 colour map must be changed to cause colours to flash." This appears to suggest
764 that the palette registers must be updated upon the flash counter - read and
765 written by OSBYTE &C1 (193) - reaching zero and that some way of changing the
766 colour pairs to be any combination of colours might be possible, instead of
767 having colour complements as pairs.
768
769 It is conceivable that the interrupt code responsible does the simple thing
770 and merely inverts the current values for any logical colours (LC) for which
771 the associated physical colour (as supplied as the second parameter to the VDU
772 19 call) has the top bit of its four bit value set. These top bits are not
773 recorded in the palette registers but are presumably recorded separately and
774 used to build bitmaps as follows:
775
776 LC 2 colour 4 colour 16 colour 4-bit value for inversion
777 -- -------- -------- --------- -------------------------
778 0 00010001 00010001 00010001 1, 1, 1
779 1 01000100 00100010 00010001 4, 2, 1
780 2 01000100 00100010 4, 2
781 3 10001000 00100010 8, 2
782 4 00010001 1
783 5 00010001 1
784 6 00100010 2
785 7 00100010 2
786 8 01000100 4
787 9 01000100 4
788 10 10001000 8
789 11 10001000 8
790 12 01000100 4
791 13 01000100 4
792 14 10001000 8
793 15 10001000 8
794
795 Inversion value calculation:
796
797 2 colour formula: 1 << (colour * 2)
798 4 colour formula: 1 << colour
799 16 colour formula: 1 << ((colour & 2) + ((colour & 8) * 2))
800
801 For example, where logical colour 0 has been mapped to a physical colour in
802 the range 8 to 15, a bitmap of 00010001 would be chosen as its contribution to
803 the inversion operation. (The lower three bits of the physical colour would be
804 used to set the underlying colour information affected by the inversion
805 operation.)
806
807 An operation in the interrupt code would then combine the bitmaps for all
808 logical colours in 2 and 4 colour modes, with the 16 colour bitmaps being
809 combined for groups of logical colours as follows:
810
811 Logical colours
812 ---------------
813 0, 2, 8, 10
814 4, 6, 12, 14
815 5, 7, 13, 15
816 1, 3, 9, 11
817
818 These combined bitmaps would be EORed with the existing palette register
819 values in order to perform the value inversion necessary to produce the
820 flashing effect.
821
822 Thus, in the VDU 19 operation, the appropriate inversion value would be
823 calculated for the logical colour, and this value would then be combined with
824 other inversion values in a dedicated memory location corresponding to the
825 colour's group as indicated above. Meanwhile, the palette channel values would
826 be derived from the lower three bits of the specified physical colour and
827 combined with other palette data in dedicated memory locations corresponding
828 to the palette registers.
829
830 Interestingly, although flashing colours on the BBC Micro are controlled by
831 toggling bit 0 of the &FE20 control register location for the Video ULA, the
832 actual colour inversion is done in hardware.
833
834 Enhancement: Palette Definition Lists
835 -------------------------------------
836
837 It can be useful to redefine the palette in order to change the colours
838 available for a particular region of the screen, particularly in modes where
839 the choice of colours is constrained, and if an increased colour depth were
840 available, palette redefinition would be useful to give the illusion of more
841 than 16 colours in MODE 2. Traditionally, palette redefinition has been done
842 by using interrupt-driven timers, but a more efficient approach would involve
843 presenting lists of palette definitions to the ULA so that it can change the
844 palette at a particular display line.
845
846 One might define a palette redefinition list in a region of memory and then
847 communicate its contents to the ULA by writing the address and length of the
848 list, along with the display line at which the palette is to be changed, to
849 ULA registers such that the ULA buffers the list and performs the redefinition
850 at the appropriate time. Throughput/bandwidth considerations might impose
851 restrictions on the practical length of such a list, however.
852
853 Enhancement: Display Synchronisation Interrupts
854 -----------------------------------------------
855
856 When completing each scanline of the display, the ULA could trigger an
857 interrupt. Since this might impact system performance substantially, the
858 feature would probably need to be configurable, and it might be sufficient to
859 have an interrupt only after a certain number of display lines instead.
860 Permitting the CPU to take action after eight lines would allow palette
861 switching and other effects to occur on a character row basis.
862
863 The ULA provides an interrupt at the end of the display period, presumably so
864 that software can schedule updates to the screen, avoid flickering or tearing,
865 and so on. However, some applications might benefit from an interrupt at, or
866 just before, the start of the display period so that palette modifications or
867 similar effects could be scheduled.
868
869 Enhancement: Palette-Free Modes
870 -------------------------------
871
872 Palette-free modes might be defined where bit values directly correspond to
873 the red, green and blue channels, although this would mostly make sense only
874 for modes with depths greater than the standard 4 bits per pixel, and such
875 modes would require more memory than MODE 2 if they were to have an acceptable
876 resolution.
877
878 Enhancement: Display Suspend
879 ----------------------------
880
881 Especially when writing to the screen memory, it could be beneficial to be
882 able to suspend the ULA's access to the memory, instead producing blank values
883 for all screen pixels until a program is ready to reveal the screen. This is
884 different from palette blanking since with a blank palette, the ULA is still
885 reading screen memory and translating its contents into pixel values that end
886 up being blank.
887
888 This function is reminiscent of a capability of the ZX81, albeit necessary on
889 that hardware to reduce the load on the system CPU which was responsible for
890 producing the video output. By allowing display suspend on the Electron, the
891 performance benefit would be derived from giving the CPU full access to the
892 memory bandwidth.
893
894 The region blanking feature mentioned above could be implemented using this
895 enhancement instead of employing palette blanking for the affected lines of
896 the display.
897
898 Enhancement: Memory Filling
899 ---------------------------
900
901 A capability that could be given to an enhanced ULA is that of permitting the
902 ULA to write to screen memory as well being able to read from it. Although
903 such a capability would probably not be useful in conjunction with the
904 existing read operations when producing a screen display, and insufficient
905 bandwidth would exist to do so in high-bandwidth screen modes anyway, the
906 capability could be offered during a display suspend period (as described
907 above), permitting a more efficient mechanism to rapidly fill memory with a
908 predetermined value.
909
910 This capability could also support block filling, where the limits of the
911 filled memory would be defined by the position and size of a screen area,
912 although this would demand the provision of additional registers in the ULA to
913 retain the details of such areas and additional logic to control the fill
914 operation.
915
916 Enhancement: Region Filling
917 ---------------------------
918
919 An alternative to memory writing might involve indicating regions using
920 additional registers or memory where the ULA fills regions of the screen with
921 content instead of reading from memory. Unlike hardware sprites which should
922 realistically provide varied content, region filling could employ single
923 colours or patterns, and one advantage of doing so would be that the ULA need
924 not access memory at all within a particular region.
925
926 Regions would be defined on a row-by-row basis. Instead of reading memory and
927 blitting a direct representation to the screen, the ULA would read region
928 definitions containing a start column, region width and colour details. There
929 might be a certain number of definitions allowed per row, or the ULA might
930 just traverse an ordered list of such definitions with each one indicating the
931 row, start column, region width and colour details.
932
933 One could even compress this information further by requiring only the row,
934 start column and colour details with each subsequent definition terminating
935 the effect of the previous one. However, one would also need to consider the
936 convenience of preparing such definitions and whether efficient access to
937 definitions for a particular row might be desirable. It might also be
938 desirable to avoid having to prepare definitions for "empty" areas of the
939 screen, effectively making the definition of the screen contents employ
940 run-length encoding and employ only colour plus length information.
941
942 One application of region filling is that of simple 2D and 3D shape rendering.
943 Although it is entirely possible to plot such shapes to the screen and have
944 the ULA blit the memory contents to the screen, such operations consume
945 bandwidth both in the initial plotting and in the final transfer to the
946 screen. Region filling would reduce such bandwidth usage substantially.
947
948 This way of representing screen images would make certain kinds of images
949 unfeasible to represent - consider alternating single pixel values which could
950 easily occur in some character bitmaps - even if an internal queue of regions
951 were to be supported such that the ULA could read ahead and buffer such
952 "bandwidth intensive" areas. Thus, the ULA might be better served providing
953 this feature for certain areas of the display only as some kind of special
954 graphics window.
955
956 Enhancement: Hardware Sprites
957 -----------------------------
958
959 An enhanced ULA might provide hardware sprites, but this would be done in an
960 way that is incompatible with the standard ULA, since no &FE*X locations are
961 available for allocation. To keep the facility simple, hardware sprites would
962 have a standard byte width and height.
963
964 The specification of sprites could involve the reservation of 16 locations
965 (for example, &FE20-F) specifying a fixed number of eight sprites, with each
966 location pair referring to the sprite data. By limiting the ULA to dealing
967 with a fixed number of sprites, the work required inside the ULA would be
968 reduced since it would avoid having to deal with arbitrary numbers of sprites.
969
970 The principal limitation on providing hardware sprites is that of having to
971 obtain sprite data, given that the ULA is usually required to retrieve screen
972 data, and given the lack of memory bandwidth available to retrieve sprite data
973 (particularly from multiple sprites supposedly at the same position) and
974 screen data simultaneously. Although the ULA could potentially read sprite
975 data and screen data in alternate memory accesses in screen modes where the
976 bandwidth is not already fully utilised, this would result in a degradation of
977 performance.
978
979 Enhancement: Additional Screen Mode Configurations
980 --------------------------------------------------
981
982 Alternative screen mode configurations could be supported. The ULA has to
983 produce 640 pixel values across the screen, with pixel doubling or quadrupling
984 employed to fill the screen width:
985
986 Screen width Columns Scaling Depth Bytes
987 ------------ ------- ------- ----- -----
988 640 80 x1 1 80
989 320 40 x2 1, 2 40, 80
990 160 20 x4 2, 4 40, 80
991
992 It must also use at most 80 byte-sized memory accesses to provide the
993 information for the display. Given that characters must occupy an 8x8 pixel
994 array, if a configuration featuring anything other than 20, 40 or 80 character
995 columns is to be supported, compromises must be made such as the introduction
996 of blank pixels either between characters (such as occurs between rows in MODE
997 3 and 6) or at the end of a scanline (such as occurs at the end of the frame
998 in MODE 3 and 6). Consider the following configuration:
999
1000 Screen width Columns Scaling Depth Bytes Blank
1001 ------------ ------- ------- ----- ------ -----
1002 208 26 x3 1, 2 26, 52 16
1003
1004 Here, if the ULA can triple pixels, a 26 column mode with either 2 or 4
1005 colours could be provided, with 16 blank pixel values (out of a total of 640)
1006 generated either at the start or end (or split between the start and end) of
1007 each scanline.
1008
1009 Enhancement: Character Attributes
1010 ---------------------------------
1011
1012 The BBC Micro MODE 7 employs something resembling character attributes to
1013 support teletext displays, but depends on circuitry providing a character
1014 generator. The ZX Spectrum, on the other hand, provides character attributes
1015 as a means of colouring bitmapped graphics. Although such a feature is very
1016 limiting as the sole means of providing multicolour graphics, in situations
1017 where the choice is between low resolution multicolour graphics or high
1018 resolution monochrome graphics, character attributes provide a potentially
1019 useful compromise.
1020
1021 For each byte read, the ULA must deliver 8 pixel values (out of a total of
1022 640) to the video output, doing so by either emptying its pixel buffer on a
1023 pixel per cycle basis, or by multiplying pixels and thus holding them for more
1024 than one cycle. For example for a screen mode having 640 pixels in width:
1025
1026 Cycle: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1027 Reads: B B
1028 Pixels: 0 1 2 3 4 5 6 7 0 1 2 3 4 5 6 7
1029
1030 And for a screen mode having 320 pixels in width:
1031
1032 Cycle: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1033 Reads: B
1034 Pixels: 0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7
1035
1036 However, in modes where less than 80 bytes are required to generate the pixel
1037 values, an enhanced ULA might be able to read additional bytes between those
1038 providing the bitmapped graphics data:
1039
1040 Cycle: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1041 Reads: B A
1042 Pixels: 0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7
1043
1044 These additional bytes could provide colour information for the bitmapped data
1045 in the following character column (of 8 pixels). Since it would be desirable
1046 to apply attribute data to the first column, the initial 8 cycles might be
1047 configured to not produce pixel values.
1048
1049 For an entire character, attribute data need only be read for the first row of
1050 pixels for a character. The subsequent rows would have attribute information
1051 applied to them, although this would require the attribute data to be stored
1052 in some kind of buffer. Thus, the following access pattern would be observed:
1053
1054 Reads: A B _ B _ B _ B _ B _ B _ B _ B ...
1055
1056 In modes 3 and 6, the blank display lines could be used to retrieve attribute
1057 data:
1058
1059 Reads (blank): A _ A _ A _ A _ A _ A _ A _ A _ ...
1060 Reads (active): B _ B _ B _ B _ B _ B _ B _ B _ ...
1061 Reads (active): B _ B _ B _ B _ B _ B _ B _ B _ ...
1062 ...
1063
1064 See below for a discussion of using this for character data as well.
1065
1066 A whole byte used for colour information for a whole character would result in
1067 a choice of 256 colours, and this might be somewhat excessive. By only reading
1068 attribute bytes at every other opportunity, a choice of 16 colours could be
1069 applied individually to two characters.
1070
1071 Cycle: 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
1072 Reads: B A B -
1073 Pixels: 0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7 0 0 1 1 2 2 3 3 4 4 5 5 6 6 7 7
1074
1075 Further reductions in attribute data access, offering 4 colours for every
1076 character in a four character block, for example, might also be worth
1077 considering.
1078
1079 Consider the following configurations for screen modes with a colour depth of
1080 1 bit per pixel for bitmap information:
1081
1082 Screen width Columns Scaling Bytes (B) Bytes (A) Colours Screen start
1083 ------------ ------- ------- --------- --------- ------- ------------
1084 320 40 x2 40 40 256 &5300
1085 320 40 x2 40 20 16 &5580 -> &5500
1086 320 40 x2 40 10 4 &56C0 -> &5600
1087 208 26 x3 26 26 256 &62C0 -> &6200
1088 208 26 x3 26 13 16 &6460 -> &6400
1089
1090 Enhancement: Text-Only Modes using Character and Attribute Data
1091 ---------------------------------------------------------------
1092
1093 In modes 3 and 6, the blank display lines could be used to retrieve character
1094 and attribute data instead of trying to insert it between bitmap data accesses,
1095 but this data would then need to be retained:
1096
1097 Reads: A C A C A C A C A C A C A C A C ...
1098 Reads: B _ B _ B _ B _ B _ B _ B _ B _ ...
1099
1100 Only attribute (A) and character (C) reads would require screen memory
1101 storage. Bitmap data reads (B) would involve either accesses to memory to
1102 obtain character definition details or could, at the cost of special storage
1103 in the ULA, involve accesses within the ULA that would then free up the RAM.
1104 However, the CPU would not benefit from having any extra access slots due to
1105 the limitations of the RAM access mechanism.
1106
1107 A scheme without caching might be possible. The same line of memory addresses
1108 might be visited over and over again for eight display lines, with an index
1109 into the bitmap data being incremented from zero to seven. The access patterns
1110 would look like this:
1111
1112 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 0)
1113 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 1)
1114 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 2)
1115 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 3)
1116 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 4)
1117 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 5)
1118 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 6)
1119 Reads: C B C B C B C B C B C B C B C B ... (generate data from index 7)
1120
1121 The bandwidth requirements would be the sum of the accesses to read the
1122 character values (repeatedly) and those to read the bitmap data to reproduce
1123 the characters on screen.
1124
1125 Enhancement: MODE 7 Emulation using Character Attributes
1126 --------------------------------------------------------
1127
1128 If the scheme of applying attributes to character regions were employed to
1129 emulate MODE 7, in conjunction with the MODE 6 display technique, the
1130 following configuration would be required:
1131
1132 Screen width Columns Rows Bytes (B) Bytes (A) Colours Screen start
1133 ------------ ------- ---- --------- --------- ------- ------------
1134 320 40 25 40 20 16 &5ECC -> &5E00
1135 320 40 25 40 10 4 &5FC6 -> &5F00
1136
1137 Although this requires much more memory than MODE 7 (8500 bytes versus MODE
1138 7's 1000 bytes), it does not need much more memory than MODE 6, and it would
1139 at least make a limited 40-column multicolour mode available as a substitute
1140 for MODE 7.
1141
1142 Using the text-only enhancement with caching of data or with repeated reads of
1143 the same character data line for eight display lines, the storage requirements
1144 would be diminished substantially:
1145
1146 Screen width Columns Rows Bytes (C) Bytes (A) Colours Screen start
1147 ------------ ------- ---- --------- --------- ------- ------------
1148 320 40 25 40 20 16 &7A94 -> &7A00
1149 320 40 25 40 10 4 &7B1E -> &7B00
1150 320 40 25 40 5 2 &7B9B -> &7B00
1151 320 40 25 40 0 (2) &7C18 -> &7C00
1152 640 80 25 80 40 16 &7448 -> &7400
1153 640 80 25 80 20 4 &763C -> &7600
1154 640 80 25 80 10 2 &7736 -> &7700
1155 640 80 25 80 0 (2) &7830 -> &7800
1156
1157 Note that the colours describe the locally defined attributes for each
1158 character. When no attribute information is provided, the colours are defined
1159 globally.
1160
1161 Enhancement: Compressed Character Data
1162 --------------------------------------
1163
1164 Another observation about text-only modes is that they only need to store a
1165 restricted set of bitmapped data values. Encoding this set of values in a
1166 smaller unit of storage than a byte could possibly help to reduce the amount
1167 of storage and bandwidth required to reproduce the characters on the display.
1168
1169 Enhancement: High Resolution Graphics
1170 -------------------------------------
1171
1172 Screen modes with higher resolutions and larger colour depths might be
1173 possible, but this would in most cases involve the allocation of more screen
1174 memory, and the ULA would probably then be obliged to page in such memory for
1175 the CPU to be able to sensibly access it all.
1176
1177 Enhancement: Genlock Support
1178 ----------------------------
1179
1180 The ULA generates a video signal in conjunction with circuitry producing the
1181 output features necessary for the correct display of the screen image.
1182 However, it appears that the ULA drives the video synchronisation mechanism
1183 instead of reacting to an existing signal. Genlock support might be possible
1184 if the ULA were made to be responsive to such external signals, resetting its
1185 address generators upon receiving synchronisation events.
1186
1187 Enhancement: Improved Sound
1188 ---------------------------
1189
1190 The standard ULA reserves &FE*6 for sound generation and cassette input/output
1191 (with bits 1 and 2 of &FE*7 being used to select either sound generation or
1192 cassette I/O), thus making it impossible to support multiple channels within
1193 the given framework. The BBC Micro ULA employs &FE40-&FE4F for sound control,
1194 and an enhanced ULA could adopt this interface.
1195
1196 The BBC Micro uses the SN76489 chip to produce sound, and the entire
1197 functionality of this chip could be emulated for enhanced sound, with a subset
1198 of the functionality exposed via the &FE*6 interface.
1199
1200 See: http://en.wikipedia.org/wiki/Texas_Instruments_SN76489
1201 See: http://www.smspower.org/Development/SN76489
1202
1203 Enhancement: Waveform Upload
1204 ----------------------------
1205
1206 As with a hardware sprite function, waveforms could be uploaded or referenced
1207 using locations as registers referencing memory regions.
1208
1209 Enhancement: Sound Input/Output
1210 -------------------------------
1211
1212 Since the ULA already controls audio input/output for cassette-based data, it
1213 would have been interesting to entertain the idea of sampling and output of
1214 sounds through the cassette interface. However, a significant amount of
1215 circuitry is employed to process the input signal for use by the ULA and to
1216 process the output signal for recording.
1217
1218 See: http://bbc.nvg.org/doc/A%20Hardware%20Guide%20for%20the%20BBC%20Microcomputer/bbc_hw_03.htm#3.11
1219
1220 Enhancement: BBC ULA Compatibility
1221 ----------------------------------
1222
1223 Although some new ULA functions could be defined in a way that is also
1224 compatible with the BBC Micro, the BBC ULA is itself incompatible with the
1225 Electron ULA: &FE00-7 is reserved for the video controller in the BBC memory
1226 map, but controls various functions specific to the 6845 video controller;
1227 &FE08-F is reserved for the serial controller. It therefore becomes possible
1228 to disregard compatibility where compatibility is already disregarded for a
1229 particular area of functionality.
1230
1231 &FE20-F maps to video ULA functionality on the BBC Micro which provides
1232 control over the palette (using address &FE21, compared to &FE07-F on the
1233 Electron) and other system-specific functions. Since the location usage is
1234 generally incompatible, this region could be reused for other purposes.
1235
1236 Enhancement: Increased RAM, ULA and CPU Performance
1237 ---------------------------------------------------
1238
1239 More modern implementations of the hardware might feature faster RAM coupled
1240 with an increased ULA clock frequency in order to increase the bandwidth
1241 available to the ULA and to the CPU in situations where the ULA is not needed
1242 to perform work. A ULA employing a 32MHz clock would be able to complete the
1243 retrieval of a byte from RAM in only 250ns and thus be able to enable the CPU
1244 to access the RAM for the following 250ns even in display modes requiring the
1245 retrieval of a byte for the display every 500ns. The CPU could, subject to
1246 timing issues, run at 2MHz even in MODE 0, 1 and 2.
1247
1248 A scheme such as that described above would have a similar effect to the
1249 scheme employed in the BBC Micro, although the latter made use of RAM with a
1250 wider bandwidth in order to complete memory transfers within 250ns and thus
1251 permit the CPU to run continuously at 2MHz.
1252
1253 Higher bandwidth could potentially be used to implement exotic features such
1254 as RAM-resident hardware sprites or indeed any feature demanding RAM access
1255 concurrent with the production of the display image.
1256
1257 Enhancement: Multiple CPU Stacks and Zero Pages
1258 -----------------------------------------------
1259
1260 The 6502 maintains a stack for subroutine calls and register storage in page
1261 &01. Although the stack register can be manipulated using the TSX and TXS
1262 instructions, thereby permitting the maintenance of multiple stack regions and
1263 thus the potential coexistence of multiple programs each using a separate
1264 region, only programs that make little use of the stack (perhaps avoiding
1265 deeply-nested subroutine invocations and significant register storage) would
1266 be able to coexist without overwriting each other's stacks.
1267
1268 One way that this issue could be alleviated would involve the provision of a
1269 facility to redirect accesses to page &01 to other areas of memory. The ULA
1270 would provide a register that defines a physical page for the use of the CPU's
1271 "logical" page &01, and upon any access to page &01 by the CPU, the ULA would
1272 change the asserted address lines to redirect the access to the appropriate
1273 physical region.
1274
1275 By providing an 8-bit register, mapping to the most significant byte (MSB) of
1276 a 16-bit address, the ULA could then replace any MSB equal to &01 with the
1277 register value before the access is made. Where multiple programs coexist,
1278 upon switching programs, the register would be updated to point the ULA to the
1279 appropriate stack location, thus providing a simple memory management unit
1280 (MMU) capability.
1281
1282 In a similar fashion, zero page accesses could also be redirected so that code
1283 could run from sideways RAM and have zero page operations redirected to "upper
1284 memory" - for example, to page &BE (with stack accesses redirected to page
1285 &BF, perhaps) - thereby permitting most CPU operations to occur without
1286 inadvertent accesses to "lower memory" (the RAM) which would risk stalling the
1287 CPU as it contends with the ULA for memory access.
1288
1289 Such facilities could also be provided by a separate circuit between the CPU
1290 and ULA in a fashion similar to that employed by a "turbo" board, but unlike
1291 such boards, no additional RAM would be provided: all memory accesses would
1292 occur as normal through the ULA, albeit redirected when configured
1293 appropriately.
1294
1295 ULA Pin Functions
1296 -----------------
1297
1298 The functions of the ULA pins are described in the Electron Service Manual. Of
1299 interest to video processing are the following:
1300
1301 CSYNC (low during horizontal or vertical synchronisation periods, high
1302 otherwise)
1303
1304 HS (low during horizontal synchronisation periods, high otherwise)
1305
1306 RED, GREEN, BLUE (pixel colour outputs)
1307
1308 CLOCK IN (a 16MHz clock input, 4V peak to peak)
1309
1310 PHI OUT (a 1MHz, 2MHz and stopped clock signal for the CPU)
1311
1312 More general memory access pins:
1313
1314 RAM0...RAM3 (data lines to/from the RAM)
1315
1316 RA0...RA7 (address lines for sending both row and column addresses to the RAM)
1317
1318 RAS (row address strobe setting the row address on a negative edge - see the
1319 timing notes)
1320
1321 CAS (column address strobe setting the column address on a negative edge -
1322 see the timing notes)
1323
1324 WE (sets write enable with logic 0, read with logic 1)
1325
1326 ROM (select data access from ROM)
1327
1328 CPU-oriented memory access pins:
1329
1330 A0...A15 (CPU address lines)
1331
1332 PD0...PD7 (CPU data lines)
1333
1334 R/W (indicates CPU write with logic 0, CPU read with logic 1)
1335
1336 Interrupt-related pins:
1337
1338 NMI (CPU request for uninterrupted 1MHz access to memory)
1339
1340 IRQ (signal event to CPU)
1341
1342 POR (power-on reset, resetting the ULA on a positive edge and asserting the
1343 CPU's RST pin)
1344
1345 RST (master reset for the CPU signalled on power-up and by the Break key)
1346
1347 Keyboard-related pins:
1348
1349 KBD0...KBD3 (keyboard inputs)
1350
1351 CAPS LOCK (control status LED)
1352
1353 Sound-related pins:
1354
1355 SOUND O/P (sound output using internal oscillator)
1356
1357 Cassette-related pins:
1358
1359 CAS IN (cassette circuit input, between 0.5V to 2V peak to peak)
1360
1361 CAS OUT (pseudo-sinusoidal output, 1.8V peak to peak)
1362
1363 CAS RC (detect high tone)
1364
1365 CAS MO (motor relay output)
1366
1367 ÷13 IN (~1200 baud clock input)
1368
1369 ULA Socket
1370 ----------
1371
1372 The socket used for the ULA is a 3M/TexTool 268-5400 68-pin socket.
1373
1374 References
1375 ----------
1376
1377 See: http://bbc.nvg.org/doc/A%20Hardware%20Guide%20for%20the%20BBC%20Microcomputer/bbc_hw.htm
1378
1379 About this Document
1380 -------------------
1381
1382 The most recent version of this document and accompanying distribution should
1383 be available from the following location:
1384
1385 http://hgweb.boddie.org.uk/ULA
1386
1387 Copyright and licence information can be found in the docs directory of this
1388 distribution - see docs/COPYING.txt for more information.