GitLab

Retagged as 'SprExtend-1_42'

Detail:
  Page 33 of the RiscOS 3 volume 3 Programmers manual states that the
  active point can be set from the sprite's mask. This was never
  implemented but it is very useful as it means that the active point
  is no longer hardwired into programs but dependant on the sprite
  used.

  The obvious problem is ptr_write. At present the active point is at
  the top left. If you run !ptr_test and run the pointer over the
  'standard active point position' writable icons from the bottom you
  will see that the pointer changes to an I with its top at the bottom
  of the writable icon. If you move the pointer until the top of the I
  is just inside the upper writable icon of the 2 and click, the focus
  goes into the upper writable - not what you'd expect.

  Similarly it's difficult to place the caret between the 2 lls in
  will. The bottom two icons shows the difference with the active
  point in the correct position in the I beam.

  You could change the wimp to put the caret in the correct position
  like I have in the bottom 2 writables of !Ptr_test however this
  would have the wrong active point position if the mode had
  rectangular y pixels and it wouldn't fix all instances where
  programmers had set ther own values.

  It also means that themes can change the pointer active point as
  well as the shape.

  The routine added copes with old and new type sprites of any bit depth though
  OS_SpriteOp 36 has a 8bit limit.
Admin:
  Tested on Iyonix 5.17.
Author:
  Colin Granville

Version 1.42. Tagged as 'SprExtend-1_42'

Replaced use of LSL (et al) macros with the new myLSL macros to avoid
potential clash with new pseudo-opcodes

Admin: requires HdrSrc 1.79 or later.

Version 1.41. Tagged as 'SprExtend-1_41'

The normalisation was wrong, leading to very dark images.
There was no code to deal with the 8bpp grey modes, leading to corruption.
This is the monochrome version of the same bug fixed for colour in 1.32.

Version 1.40. Tagged as 'SprExtend-1_40'

Fix bug in JPEG vetting, tag 0xC4 was being taken as an SOF and quitting the tag scanning loop, which in turn rejected the JPEG as 0xC4 isn't a supported SOF type. In fact, it's not even an SOF type at all. See new IS_AN_SOF macro.

Fix bug leaving JPEG file handle open when unknown SOF or entropy encoding encountered (spotted testing the above).

Version 1.39. Tagged as 'SprExtend-1_39'

Fold RISC OS specific colour conversion routines into rojpeg.
Remove jcconv, this also highlighted there were some more unused functions.
Should make move to IJG version 8 even easier.

Version 1.38. Tagged as 'SprExtend-1_38'

Aside from moving blocks around and adding comments, decisions about numcomponents use switch/case statements to allow CMYK support in the future.
Split out (some of the) Huffman decoding to allow arithmetic support in the future.
Moved memcpy() and memset() into CSupport.
Tested in all colour depths with 1x1; 1x2; 2x1; 2x2 sampling & ChangeFSI.
Tighten up check on cinfo being NULL (was previously only checked in the colour decoder, not the monochrome).

Version 1.37. Tagged as 'SprExtend-1_37'

When clear, r2 and r3 are preserved per PRM documentation.
Also BIC'd out bits 7-10 of r0 on return so they can be reused in future, previous version would always have returned &C in those bits.

Version 1.36. Tagged as 'SprExtend-1_36'

Setting bit 1 of r0 in JPEG_FileInfo and JPEG_Info will now return the SOF type of the JPEG in question in bits 3-6 of the info flags in r0 on exit.
Took this opportunity to refactor find_image_dims to take a structure pointer rather than lots of individual pointers (CFSI-JPEG ammended to match). Untangled the return codes a bit so it is only necessary to inspect r0 on exit from C to determine the outcome. Values of 1 & 2 are translated in the assembler code into more specific error messages than just "Incomplete or corrupt JPEG".
Fix bug where unknown density marker would leave the file open in JPEG_FileInfo.
Don't bother ADRL'ing r1 for error messages that don't have a %0 in them.
Added all the input & output flag definitions to the exported header file so other callers can use them too.
More functionality, fewer bugs, about 200 bytes smaller too.

Version 1.35. Tagged as 'SprExtend-1_35'

In find_image_dims() the while loop is only exited when SOF0 or SOF1 (the two supported ones) is encountered, the other 14 possible combinations got stuck in a loop because OS_GBPB doesn't return an error at EOF.
Check added for OS_GBPB returning nothing, plus an exit when a SOS or EOI tag is hit for the memory only check (it's not possible to know where the end truly is in that case.
jcconv: warning supressed.
SWIs: comment corrected.
Tested with a bogus SOF, now gives an error message instead.

Version 1.34. Tagged as 'SprExtend-1_34'

There's no need for the C code to know which pool of memory it's using, so simplify it to just use malloc() and realloc(), the assembler implementation can then select whether the RMA or a dynamic area is in use based on the 'dynamicarea' switch.
Be more careful about area_number when it is zero, since we use that to denote the area is not yet set up but area zero is a valid dynamic area number!
Trim out some unused workspace.
Variable ccompiler_errptr had '# 0' bytes assigned, resulting in writes off the end of the workspace.
Retire DUMPVAL[N]() macro.
Remove M_SOF9 as one of the SOF markers accepted - this is allocated to arithmetic JPEGs which aren't currently supported.
Save a few words by: changing arbitrary constants (eg. safety margins) to be a power of two so they can be register immediates; numerous loads and stores of R12 relative variables changed to not use ADRL; jpeg_fetchroutine similarly simplified.

Version 1.33. Tagged as 'SprExtend-1_33'

The exit_c function was repeatedly being called with the line number argument and error code the wrong way round and since only bits 0-2 of the error code are inspected on exit any fatal exits from the ported JPEG library code would either report the wrong error or fail silently (depending on the line number) - example a 12 bits per sample JPEG would be rendered without fault into a screen full of random memory garbage.
Assembling/compiling with the 'jpeg' switch off didn't work, fixed, and also now zeros out the SWI decode values in the module header.
Moved include of 'yuvtabs.s' into the colour conversion code.
Deleted panic_exit, it did nothing useful but trace a value then exit, so it now just calls exit() like everyone else.
Removed excess call to find_image_dims(), this was needed just to get the workspace requirements when the JPEG had already been sized up by asm_find_image_dims() earlier. Side effect of removing the pilot JPEG code earlier is that this call can no...

Function asm_get_image_dims returns width in r2, was using r0, so version 1.30 would leave shadows in !Draw. Changed to use r2.

Version 1.31. Tagged as 'SprExtend-1_31'

During the original development (referred to in the docs as Sprite Extend 2) JPEG data was briefly wrapped inside a sprite file before the JPEG_ swis were introduced. Within SpriteExtend some code still remained to fake a mode 20 sprite header with the JPEG output sized adjusted as though it was 4bpp, then later undo the faking by checking for a mode 20 sprite with is_it_jpeg set.
This version removes all of the code looking out for pilot JPEGs, and rationalises the fake sprite header to just report the true X/Y size.
Fix bug where r4 would be corrupted on exit from JPEG_PlotScaled if an error occurred, and generally rationalised the exit cases a bit more sensibly.
Made the JPEG_ swi despatch a bit more simple to follow.
Corrected a comment in the NSM_bpptable.
Deleted unused 'ecfspace' from the workspace.
Hid the 'macroword' from the C code as it's now an assembler only value.
Cut the initial comments from c/putscaled and placed in Docs.
Cut the debug tracing code (a notable %age) from c/putscaled and placed in tracing support file.
Simplified the entry to putscaled_compiler() now we don't have to look out for pilot format JPEGs within sprites any more.
Sprinkled in some block comments to make navigating the c/putscaled file a bit easier.

Version 1.30. Tagged as 'SprExtend-1_30'

Retagged as SprExtend-1_29.

Had erroneously been switched out in version 1.27 clean up, when it is needed. Comment added to denote it's a fall through of the table above it.

Version 1.29. Tagged as 'SprExtend-1_29'

Since 2001 this module has been using objasm, but still using the C compiler to output (via -S) assembly code which is then passed through 'sed' and included via GET into the assembler code.
Changed to compile the C code into object files, and use the linker to patch up the unresolved references - this has required some reordering of the C code to make sure static functions now appear in the object file, and some reordering of the header files to ensure both the module and the CFSI version still compile without conflicts.
Minor fix to 'swiv2', the label was in the wrong place for _swi() though it's not used in this module so didn't affect the _swix() case.
Makefile cut down a bit. Unixify some path names.
Test code moved to 'Test/c' directory.
Tested in both debug and non debug cases via softloaded module. Oddly, it's 4 bytes longer than the previous version.

Version 1.28. Tagged as 'SprExtend-1_28'

Deleted c.SpriteOp to untangle it from c.rojpeg to draw boundary around JPEG code.
Moved all the OS_SpriteOp code out of SprExtend into a new file SprOp, so the main file just contains the usual macros/includes/module heading.
Added new 'debugso' switch for debugging SpriteOps, was previously using 'debuggs' which is supposed to be for JPEG leading to reams of confusing debug.
TestForMaskAtDepth was rejecting 8bpp sprites with masks due to testing against log2bpp (4) instead of sprite type (5).
NSM_bpptable re-encoded in bytes rather than words, accesses ammended accordingly.
Took fresh copy of swiv2 from RISCOS_Lib revision 4.7.
SpriteV interception made a bit simpler (& faster).
Fixes to validate sprite (SpriteOp 17):
 Was looking at the wrong bits of the mode word to test for old mode words.
 Added new test for DPI values being non zero.
 Algorithm more closely follows PRM.

Version 1.27. Tagged as 'SprExtend-1_27'

No code change - not tagged.

2000 lines of unloved code from 1993 deleted, binary still the same afterwards, so should be same as 1.25.

Version 1.26. Tagged as 'SprExtend-1_26'

1 pixel wide JPEGs:
The diffuse dithering code considers all lines as 1-N-1 where N can be zero and the first and last pixel are dealt with specially because there's no data to scatter beyond the edge.
With a 1 pixel wide JPEG the dithering code ended up with a -ve number in the X counter and wandered off the end of memory. Fix just moves a SUB up a few lines and turns it into a SUBS so that the 1 wide case becomes 0-N-1 with N=0.
1 pixel high sprites:
With output switched to a 1 pixel high sprite an abort would quickly follow because the read of mode variables had a backdoor check to workaround a bug in RISC OS 2.00 which returned a duff value for YWindLimit in MODE 22. As the definition of MODE 22 got changed to 768x288 in RISC OS 3.xx (and the YWindLimit bug fixed) this code now just falsely reported a 1 pixel high sprite as being 1024 high and the blitter splattered all over memory it shouldn't.
Result:
Loss of RISC OS 2.00 compatibility, but can now load J...

Trim 200+ lines of changes dating back to 1987, combine with non duplicates in the 'BlackLog'.
Use 'offsetof' macro in genjhdr to avoid warnings about uninitialised variables, rename from makejhdr so all the utils are clearly demarked.
Add throwback to MkCFSI-j.
Replace reason codes passed to SWIs with textual equivalents from the respective header files.
Speed up SWI despatch a bit by making the first few debug instructions depend on 'debuggs'.
Qualify all of the debug switches with the global 'debug' one.
Eliminate single use of 'pullx' macro.
Make do_branch() take an unsigned opcode so 50% of the branches don't result in implied cast warnings.
Added an UNUSED() macro to reduce a few of the legitimate warnings (the remainder are either in the original JPEG code or ones that it is worth checking the logic of.
Unixify one last filename.

Version 1.24. Tagged as 'SprExtend-1_24'

Sources/PutScaled:
 Sync comment with C code.
Sources/SprExtend:
 Debug switch 'debuggs' had been left on, which left in some unnecessary code, but turning it off pulled in code with a stack imbalance. Now off.
 Ensured a terminator at the end of the RMEnsure.
 Colour translation tables passed to SetPointerShape that were in top bit set addresses were being interpreted as though no tt was supplied. Now tested against zero to match docs.
Sources/SprTrans:
 Switches pre 0.60 collapsed.
Sources/diffuse:
 The C code was corrupting r12 workspace pointer so the check of the flag 'ctrans_recent' sometimes failed leading to a duff table lookup.
 Small optimisation to checking of guard word.
c/PutScaled:
 Change paths to be Unix friendly.
rojpeg:
 Put assembler_panic in the header file to shut the compiler up.
3x unused files deleted.

Version 1.23. Tagged as 'SprExtend-1_23'

Removed braces selecting code from 0.47 0.50 and 0.55, this is over 80
versions ago and just confusing.
Binary the same, not tagged.

The code generator could output TST instructions with immediate constants not generated in the ARM approved way, while this isn't currently a problem the generator is changed to encode constants less than 256 as an immediate with no shift.
Use of LDR rn, Rd,[Rn] would produce
       LDR Rd,[Rn],#0
which is unpredictable when Rd=Rn. Swapped use of OFFSET0 to OFFSET(0) so
an indexed load is used instead.
Removed the explicit 'S' from TST use, since the TST macro includes 'S' already.
Delete h/swis (held centrally).
Delete h/kernel (held centrally).
Delete hdr/Modes (not used).

Version 1.22. Tagged as 'SprExtend-1_22'

Previous changes were untagged, but incrementing the version number gives
us a tag we can associate with the official RISC OS 5.16 release.

Version 1.21. Tagged as 'SprExtend-1_21'

Detail:
  Defines UXTH and SXTH as macros. This means we can assemble the output of
  cc -S -cpu 6 even using current versions of objasm that are limited to the
  v5TEJ instruction set. It is expected that objasm will ignore these macros
  if/when it gains v6 instruction set support.
Admin:
  Tested - it works.

Detail:
  The build utilities utils.makejhdr, utils.genpalette and utils.genyuvtabs,
  compiled at build time, are now built using the HostTools shared makefile,
  and so will execute on any RISC OS build host even if the default compiler
  options are to target an incompatible ARM architecture. Also now sets the
  throwback flag when doing builds via the Mk obey file.
Admin:
  Tested using an APCS-R build on an Iyonix.

Version 1.20. Not tagged

Detail:
  Uses suffixed file extensions for compatiblity with both objasm and asasm.
Admin:
  Supplied by Peter Naulls, tested at ROOL

Version 1.20. Not tagged

Detail:
  The use of ARMv4 SMULL instructions is now switched based upon the
  Hdr:CPU.Arch header file. This means that in IOMD ROM builds, only pre-v4
  instructions are used, so it will function on all required target platforms.
  Tungsten and OMAP3 builds will continue to use SMULL.
Admin:
  Inspected binary to ensure correct behaviour. Requires HdrSrc 1.67 or
  1.62/4.162.2.4.

Version 1.20. Tagged as 'SprExtend-1_20'

Removed local definition of XOS_CLI and XColourTrans_SelectTable, replaced
with header file.

Delete swiv2_old (moved to _old directory)

Changed to use {TRUE} and {FALSE} built in objasm variables.

Label dividebyzero fall through had code in the way for debug = {TRUE}.

Trimmed some dead code and definitions.

Stopped thrashing OS_CLI every time a JPEG plot is required to check
ColourTrans version, this is now done once at startup and a flag kept
instead.

Author: Rob Sprowson

Version 1.19. Tagged as 'SprExtend-1_19'

Detail:
  Creates install directories if not already present before doing the
  copy operation.
Admin:
  Will work.

Version 1.18. Tagged as 'SprExtend-1_18'

Detail:
  cache_sprite_doesnt_exist_error routine in MsgCode failed to take into
  account the fact that the messages file might be tokenised. Calling
  MessageTrans_Lookup with R2=0 gives you the length of the unexpanded
  message in R3. Now fixed to work in a better way.
  Also rename fp to r11 in swiv2 file to avoid assembler warnings.
Admin:
  Soft-loaded module tested. Problem seems to be solved.


Version 1.17. Tagged as 'SprExtend-1_17'

C compiler that spots signed divisions of known-positive quantities (eg in
this case (<char>*254)/100), and does them as unsigned instead.

Version 1.16. Tagged as 'SprExtend-1_16'

First pass check now only looks for a SOI tag,not bothering looking
for the "JFIF" text: this lets both JFIF and EXIF files through the
first gate.
Next,improvements to the find_image_dimensions routine to properly
look for tags rather than assuming tag APP0 is always at the start of
the file as it is in ".jpg" files.Again - EXIF files are now understood
as a side effect,plus a nice speed up for when skipping unknown tags.So
rather than loading lots of unused 1k blocks it just sets the file
pointer forwards to the required point.
Tested with a handful of specially contrived EXIF files plus some frommy
Panasonic camera,plus retested the JPEGs in the ChangeFSI test directory
to check that progressive JPEGs are still correctly faulted.
Yay - !Paint and !Draw can now load digital camera piccies!

Version 1.15. Tagged as 'SprExtend-1_15'

* Transformed sprite plotting optimised to not read the screen memory when
  it could be simply avoided - greatly enhances performance on Iyonix.
* Decoding of 1x2 sampled JPEGs fixed.

Version 1.14. Tagged as 'SprExtend-1_14'

Replaced a couple of STMs in error recovery code with code sequences that don't use UNPREDICTABLE (and XScale-unsafe) instructions.

Version 1.13. Tagged as 'SprExtend-1_13'