The only reason this even existed was that ZDoom's original VC projects used __fastcall. The CMake generated project do not, they stick to __cdecl.
Since no performance gain can be seen by using __fastcall the best course of action is to just remove all traces of it from the source and forget that it ever existed.
* scalex_top
* scaley_top
* scalex_mid
* scaley_mid
* scalex_bottom
* scalex_bottom
- Added sidedef versions of the linedef flags wrapmidtex and clipmidtex (via
UDMF; names are the same). If the flag is set on the line, it applies to
both sides. Otherwise, each side can control them individually.
SVN r1645 (trunk)
so that all files are included by a central one instead of compiling
each one separately. This speeds up the compilation process by 25%
when doing a complete rebuild in Visual C.
- Cleaned up more header dependencies.
SVN r1226 (trunk)
registers AMD64 provides, this routine still needs to be written as self-
modifying code for maximum performance. The additional registers do allow
for further optimization over the x86 version by allowing all four pixels
to be in flight at the same time. The end result is that AMD64 ASM is about
2.18 times faster than AMD64 C and about 1.06 times faster than x86 ASM.
(For further comparison, AMD64 C and x86 C are practically the same for
this function.) Should I port any more assembly to AMD64, mvlineasm4 is the
most likely candidate, but it's not used enough at this point to bother.
Also, this may or may not work with Linux at the moment, since it doesn't
have the eh_handler metadata. Win64 is easier, since I just need to
structure the function prologue and epilogue properly and use some
assembler directives/macros to automatically generate the metadata. And
that brings up another point: You need YASM to assemble the AMD64 code,
because NASM doesn't support the Win64 metadata directives.
- Added an SSE version of DoBlending. This is strictly C intrinsics.
VC++ still throws around unneccessary register moves. GCC seems to be
pretty close to optimal, requiring only about 2 cycles/color. They're
both faster than my hand-written MMX routine, so I don't need to feel
bad about not hand-optimizing this for x64 builds.
- Removed an extra instruction from DoBlending_MMX, transposed two
instructions, and unrolled it once, shaving off about 80 cycles from the
time required to blend 256 palette entries. Why? Because I tried writing
a C version of the routine using compiler intrinsics and was appalled by
all the extra movq's VC++ added to the code. GCC was better, but still
generated extra instructions. I only wanted a C version because I can't
use inline assembly with VC++'s x64 compiler, and x64 assembly is a bit
of a pain. (It's a pain because Linux and Windows have different calling
conventions, and you need to maintain extra metadata for functions.) So,
the assembly version stays and the C version stays out.
- Removed all the pixel doubling r_detail modes, since the one platform they
were intended to assist (486) actually sees very little benefit from them.
- Rewrote CheckMMX in C and renamed it to CheckCPU.
- Fixed: CPUID function 0x80000005 is specified to return detailed L1 cache
only for AMD processors, so we must not use it on other architectures, or
we end up overwriting the L1 cache line size with 0 or some other number
we don't actually understand.
SVN r1134 (trunk)
drawing function. The most improvement came from being able to turn some
constant variables into immediate values with self-modifying code, but I
also managed to reorder it to make it a little faster. It's about 9% faster
than VC++'s code and 19% faster than GCC's code. That's not a huge
improvement (for VC++), but at least it's measurable.
- Removed the solid fill "optimization" from rt_shaded4cols(), because in my
testing, it didn't help any and in fact, hurt just a little bit.
- In the name of simplification, all the rt_tlate* drawers were changed to do
the translation in one step and the drawing in another. This lets me call
the untranslated drawer to do the real drawing instead of mostly duplicating
them. Performance wise, there is practically no difference from before.
SVN r771 (trunk)
- Added .txt files to the list of types (wad, zip, and pk3) that can be
loaded without listing them after -file.
- Fonts that are created by the ACS setfont command to wrap a texture now
support animated textures.
- FON2 fonts can now use their full palette for CR_UNTRANSLATED when drawn
with the hardware 2D path instead of being restricted to the game palette.
- Fixed: Toggling vid_vsync would reset the displayed fullscreen gamma to 1
on a Radeon 9000.
- Added back the off-by-one palette handling, but in a much more limited
scope than before. The skipped entry is assumed to always be at 248, and
it is assumed that all Shader Model 1.4 cards suffer from this. That's
because all SM1.4 cards are based on variants of the ATI R200 core, and the
RV250 in a Radeon 9000 craps up like this. I see no reason to assume that
other flavors of the R200 are any different. (Interesting note: With the
Radeon 9000, D3DTADDRESS_CLAMP is an invalid address mode when using the
debug Direct3D 9 runtime, but it works perfectly fine with the retail
Direct3D 9 runtime.) (Insight: The R200 probably uses bytes for all its
math inside pixel shaders. That would explain perfectly why I can't use
constants greater than 1 with PS1.4 and why it can't do an exact mapping to
every entry in the color palette.
- Fixed: The software shaded drawer did not work for 2D, because its selected
"color"map was replaced with the identitymap before being used.
- Fixed: I cannot use Printf to output messages before the framebuffer was
completely setup, meaning that Shader Model 1.4 cards could not change
resolution.
- I have decided to let remap palettes specify variable alpha values for
their colors. D3DFB no longer forces them to 255.
- Updated re2c to version 0.12.3.
- Fixed: A_Wander used threshold as a timer, when it should have used
reactiontime.
- Fixed: A_CustomRailgun would not fire at all for actors without a target
when the aim parameter was disabled.
- Made the warp command work in multiplayer, again courtesy of Karate Chris.
- Fixed: Trying to spawn a bot while not in a game made for a crashing time.
(Patch courtesy of Karate Chris.)
- Removed some floating point math from hu_scores.cpp that somebody's GCC
gave warnings for (not mine, though).
- Fixed: The SBarInfo drawbar command crashed if the sprite image was
unavailable.
- Fixed: FString::operator=(const char *) did not release its old buffer when
being assigned to the null string.
- The scanner no longer has an upper limit on the length of strings it
accepts, though short strings will be faster than long ones.
- Moved all the text scanning functions into a class. Mainly, this means that
multiple script scanner states can be stored without being forced to do so
recursively. I think I might be taking advantage of that in the near
future. Possibly. Maybe.
- Removed some potential buffer overflows from the decal parser.
- Applied Blzut3's SBARINFO update #9:
* Fixed: When using even length values in drawnumber it would cap to a 98
value instead of a 99 as intended.
* The SBarInfo parser can now accept negatives for coordinates. This
doesn't allow much right now, but later I plan to add better fullscreen
hud support in which the negatives will be more useful. This also cleans
up the source a bit since all calls for (x, y) coordinates are with the
function getCoordinates().
- Added support for stencilling actors.
- Added support for non-black colors specified with DTA_ColorOverlay to the
software renderer.
- Fixed: The inverse, gold, red, and green fixed colormaps each allocated
space for 32 different colormaps, even though each only used the first one.
- Added two new blending flags to make reverse subtract blending more useful:
STYLEF_InvertSource and STYLEF_InvertOverlay. These invert the color that
gets blended with the background, since that seems like a good idea for
reverse subtraction. They also work with the other two blending operations.
- Added subtract and reverse subtract blending operations to the renderer.
Since the ERenderStyle enumeration was getting rather unwieldy, I converted
it into a new FRenderStyle structure that lets each parameter of the
blending equation be set separately. This simplified the set up for the
blend quite a bit, and it means a number of new combinations are available
by setting the parameters properly.
SVN r710 (trunk)
