Build 3.44 is available at the location listed below

\\builds\PreRelease\nForce\Win2K\MCP1_WDM\3.44

Summary of changes since last labeled build

Bugs worked on since last labeled build

Details of changes since last labeled build

Change 743879 by jweintraub@jweintraub-home on 2003/06/05 17:03:14

VAPU-

1.) A much leaner and meaner MixInReplicatingBufferSSE (tighter code), but not much in the way of performance.

Change 743772 by jweintraub@jweintraub-home on 2003/06/05 16:26:17

VAPU-

1.) Removed the remaining 4 'fabs' functions from the VAPU. Only the CreateCenter functionality was using it, and using it is slower than just coding it directly. Now it no longer exists in any VAPU code. Tested CreateCenter, seems fine.

Note: John...the 4th order bandpass filter in the CreateCenter function is a dog, 62% of the time spent executing that code when running 6 channel with CreateCenter enabled. Without CreateCenter the driver is called 0.96% of the time, with CreateCenter enabled, 2.82%. Does this seem acceptable?

Change 743546 by jweintraub@jweintraub-home on 2003/06/05 15:20:29

VAPU-

1.) Added new routine NormalMixingStereo3DNOW to speed up 2 speaker mode (and headphone) mixing. This routine is 2.5-3.5% faster than the NormalMixingStereo routine and uses 3DNOW intrinsics. The percentages look a lot lower because 2 speaker mode does xtalk filtering and virtual rear positioning, which is much more significant (in terms of processor usage), then the final mixing routines. These other routines are not used in 4 and 6 speaker processing.
2.) A small enahancement to NormalMixing3DNOW and NormalMixingQuad3DNOW to remove some redundant code, that should make it even run a bit faster.

Change 743520 by sholmes@wilbur on 2003/06/05 15:13:29

Added class factory support so the DLL will get load when applications use CoCreate and not DirectSoundCreate. With EAX Panel I can capture direct volume writes. There is still a small problem though somewhere as I can't get hitman to run, even though its going through the DLL.

Change 743509 by sholmes@wilbur on 2003/06/05 15:09:25

Fixed bug where EAX environment was incorrect and out of range. This was causing application to fail on an Audigy.

Also modified code to allocate DSOUND via CoCreate to debug test DSOUND.

Change 743334 by jweintraub@jweintraub-home on 2003/06/05 14:20:40

VAPU-

1.) Yet another speedup to 6 channel mixing. Created NormalMixing3DNOW routine, this routine (which uses 3DNOW) is 20-22% faster than the NormalMixing routine which does not use any 3DNOW or SSE or MMX intrinsics. NormalMixingSSE was only about 8-13% faster than the original rotuine, but still slower than the 3DNOW version because it still contains 'ftol' instructions. I left NormalMixingSSE in the code but it is not referenced, only the normal version and the 3DNOW version if 3DNOW is supported.
2.) Changed NormalMixingQuadSSE routine name to NormalMixingQuad3DNOW, since it uses 3DNOW instructions and not SSE instructions.

Change 743231 by speterson@spetersondev on 2003/06/05 13:36:39

Summary: Updated I3DL2 reverb handler to pass EAX Listener flags to the reverb engine.

Change 743116 by jweintraub@jweintraub-home on 2003/06/05 12:23:05

VAPU- Oops, used the g_bSSE flag instead of the g_b3DNOW flag to enable last change. NormalMixingQuadSSE uses 3DNOW instructions not SSE or MMX as previously stated.

Change 743069 by jweintraub@jweintraub-home on 2003/06/05 12:08:47

VAPU-

1.) Added new routine, NormalMixingQuadSSE, and this uses MMX instead of SSE instructions, but the good news is I was able to get rid of the 'ftol' altogether, and subsequently show even better performance gains than with the 6 channel code of yesterday. With the 3 players WMP, mplayer2 and dsplay8 I got an 17%, 11% and 12% percent gain respectively. Even though I am only processing 2 floating point multiplies per pass instead of 4, removing the ftol with an MMX instruction showed a larger gain. Will retry 6 channel mixing with MMX to see if there is a difference next.

Change 743048 by sholmes@wilbur on 2003/06/05 11:58:17

Added basic dsound.dll shell that forwards calls to the real dsound.dll (if you rename it dsoundr.dll) to try to look at hitman parameters that are being passed. It works for my application that use DirectSoundCreate function which was linked at compile time for the application. If the application uses CoCreateInstance for doing its DirectSoundCreate call, then I get loaded but skipped which is what seems to happen.

Change 742219 by tblanchard@tblanchard-xp on 2003/06/04 18:05:48

Summary: When disabling the DSP Level Meter calls, I also disable the INVAudio polling for NVA_LEVELMETERS_CONTROL as well.

Change 742186 by jweintraub@jweintraub-home on 2003/06/04 17:50:16

VAPU

1.) Fixed the code alignment (indenting) of the main routine ProcessAllVoices.

Change 742141 by jweintraub@jweintraub-home on 2003/06/04 17:35:24

VAPU-

1.) Modified LevelMeter/Spectrum meter data so that when they are disabled the data in the internal structure is zeroed out, so that on subsequent application reads of the data, it will be zero instead of some old values.
2.) Added code to precache streams that do not have big enough SSLs to precache the full amount needed by the precaching algorithm. (i.e. Partial precache), but it turns out that some of these streams Usually 6 channel) get smaller SSLs after starting (than the first 2) which causes the streams to starve later on, so we really cant precache them the way it was intended, bastards at u$. This is currently disabled and is a work in progress.
3.) Changed some variable names to more meaningful ones.
4.) Timing changes:

Added 2 new versions of 2 routines NormalMixingSSE and MixInReplicatingBufferSSE, both are functionally identical to their non-SSE counterparts, but when these 2 routines are used (SSE bit set in CpuId register) we get a performace boost of between 6 and 14% depending on the App, on my Athlon2200XP processor. I ran VTune with 5 different players (dsplay8, mplayer2, Windows Media Player, mplay32 and Sensaura Player3D v1.55). Type of stream and number of voices should make no difference as these functions are at the very end of the pipe and are only executed once per frame and always run the same way. These 2 routines are also at the heavy end of execution when running VTune and could use even more optimization if possible (MixFrame is another, these 3 being the big three routines). Only one player dsplay8 hits GetPosition enough to register, and on dsplay8 its the worst offender of processor usage). Note: only 6 channel speaker mode is currently supported, will do other next. MixInRep will use the same routine in 4 and 6 speaker mode, but I havent hooked it in 4 speaker mode yet.

Change 742075 by tblanchard@tblanchard-xp on 2003/06/04 17:10:34

Summary: Enabled support for the ToolTip caption bar button.

BugID: 83952

Change 742051 by tblanchard@tblanchard-xp on 2003/06/04 17:00:33

Summary: Removed nForce badge logo since it is no longer used.

Change 742013 by mpereira@PEGASUS on 2003/06/04 16:42:41

Summary: A bunch of tweaks made over the last few weeks. Gradient fills, hls2rgb color selection, hide focus rects, new bitmaps, new buttons, bass, treble, reverb, etc

Bugid: none

Change 741627 by tblanchard@tblanchard-xp on 2003/06/04 14:03:29

Summary: Removed VERIFY_API() from SetForegroundWindow() as this method will return FALSE if the the window is already in the foreground.

Change 741494 by tblanchard@tblanchard-xp on 2003/06/04 12:51:31

Summary: Fixed bug where section header font size was incorrectly being computed.

Change 741316 by jkiel@JKIEL-WIN2K on 2003/06/04 11:31:27

Add scaling and do some cleanup for the rotating cube.

Change 741256 by mpereira@PEGASUS on 2003/06/04 11:14:43

Summary: bring the tray window to the foreground. Fixes cant cancel the track popup menu

Bugid: 83951

Change 740737 by tblanchard@tblanchard-xp on 2003/06/03 18:50:17

Summary: Workaround for section header fonts that are not anti-aliased.

According to MS documentation calling GetStockObject( DEFAULT_GUI_FONT ) should return Tahoma for 2000 & XP systems, but it was returning MS Sans Serif (non-TrueType font that cannot be anti-aliased).

Dialogs seem to be immune to this so I get the LOGFONT structure from one of our dialogs and modify it to be 14pt and bold.

Change 740519 by mpereira@PEGASUS on 2003/06/03 16:53:27

Summary: update to set the volume in the tray icon tool tip. updated the trayicon class to format and display a percentage. updated the volsetupview vscroll handler and mixernotificatoin to update the tooltip when the master volume (or master mute) changes.

Bugid: none

Change 740329 by sholmes@wilbur on 2003/06/03 15:30:38

Improved and added more debugging prints for 3D and EAX parameters.

Change 740290 by sholmes@wilbur on 2003/06/03 15:11:49

Very simple EAX test application. Parameters are currently set to the parameters that sound awful in the game Hitman2 while running under Win9x.

Change 740243 by jnorris@jnorris-nf420 on 2003/06/03 14:51:49

Summary: Correcting a problem with the version number of nvapu.sys.

The version number wasn't getting set. Steve fixed the problem
by moving the commented out line "# nvfilter.c" to the end of the file.

Change 739296 by tblanchard@tblanchard-xp on 2003/06/02 20:11:10

Summary: Disable DSP Level Meters when nvMixer is minimized or hidden.

Change 739156 by mpereira@PEGASUS on 2003/06/02 18:24:08

Summary: updated to set the SPDIF record enable bit if both the SPDIF record and SPDIF monitor bits are clear (SPDIF IN powered down). This fixes the issue where the Empia codec would not detect locks and unlocks correctly

Bugid: none

Change 738510 by jweintraub@jweintraub-home on 2003/06/02 14:08:45

VAPU-

1.) A quick fix for multi-voice waves that were not playing correctly in mplayer2 or WMP (DS multi-voice). We need enough data to fill 7 frames if we precache a voice before playing it, but we also need more data to process future interrupts, as we do not release the SSLs until they are played, so only 7 frames worth of data is not really enough. When playing 4 and 6 channel waves, the SSL size is a lot smaller than when just playing a stereo wave, so if there is not enough data based on the new requirements, we set the bNoPrecacheStream flag. In the future I need to figure out how to do partial precaching with the current scheme, so we wont get noise if an interrupt is missed.

Change 738331 by jweintraub@jweintraub-home on 2003/06/02 12:26:14

VAPU-

1.) Removed the 4th (additional) voice list that I added last week. There is no need for it, replaced it with a flag. This is a more solid solution anyway. This is needed before I can start the next iteration of the "data precache/missed interrupt" code enhancement, as I found yet another hole (thanks John). DS HCTs still run 100% error free. (88 pass, 0 fail)

Change 737237 by jweintraub@jweintraub-home on 2003/05/30 18:00:27

VAPU-

1.) Fixed the issue with the VAPU seemingly hanging for 5 seconds when a stream is stopped, due to SMs being enabled. The issue was when the VoiceCount goes to zero the ACI is still updating with zeros, but hammers on the LM/SM code as many as 7 times per interrupt (due to missed interrupt handling and all that nonsense)...this is not good. Are we calling LM/SM to often on every interrupt....mostly no...only when interrupts are skipped or a new voice is started. Changed code to never execute LM/SMs more than once per interrupt period, and with the first frame of data sent to the ACI (with one voice usually the only frame written).
2.) Removed the g_DisableSpectrumMeters flag and all of its logic. SMs are now are always running when an app enabled them.

Note: With LM/SM running, we are still at about 5%...with 75% of that being the LM/SM code which eventually will be normally disabled. (right Terry?)

Change 737147 by speterson@spetersondev on 2003/05/30 17:22:31

Summary: Added NVRM_TRUE64 to a couple more sources files

Reviewed by speterson

Change 736868 by speterson@spetersondev on 2003/05/30 15:46:39

Summary: Added NVRM_TRUE64 compile flag to fix build break caused by resman group

Change 736839 by jweintraub@jweintraub-home on 2003/05/30 15:34:57

VAPU-

1.) Removed 2 'fabs' calls from the LevelMeter code (replaced with x=(x<0)?-x:x). 'fabs' is very slow (because it calls something called ctrlfp which takes a lot of time - vtune). John already removed the 'fabs' from his code. This removal changed the execution time on nvmcp from 3.67% to 1.7% (with no SMs) and with SMs it now takes 5%. (Using NVAMP and NVMIXER while playing a 32kHz 16Bit Stereo wave)
2.) Removed some unused variables and code from the SpectrumMeter(LM) class.
3.) Added a shift in the final SM data do all of the bars are not so high. I may adjust this again when I compare it to the hardware version.

READ THIS: While NVMixer is loaded, whether its visible or just in the toolbar, it keeps the LMs/SMs enabled always. This is bad, not only for VAPU but for APU also, as the DSP (or VAPU) will constantly be calculating the 26 values even though the UI is not active. Terry promissed to fix this for me today!!! This should lower the VAPU total yet some more, because of the 1.7%...LMs still consume 28% of the 1.7% (plus another 17% for ftol, also part of LM code).

Change 736591 by jnorris@jnorris-nf420 on 2003/05/30 13:57:36

Summary: Should have added this file as well

Change 736585 by jnorris@jnorris-nf420 on 2003/05/30 13:56:14

Summary: Rewrote part of the filtering routine to try and improve performance.

Change 736204 by jweintraub@jweintraub-home on 2003/05/30 10:43:14

VAPU-

Got the SpectrumMeters working. We were converting the data to dB, but that was already being done higher up in the driver. They are still disabled by default as they are still way too slow and cause the system to bog down when they are enabled.

Change 735717 by jweintraub@jweintraub-home on 2003/05/29 18:15:00

Fixed an incorrect comment.

Change 735623 by jweintraub@jweintraub-home on 2003/05/29 17:22:31

VAPU-

Added code for SpectrumMeters based on John's TestVAPUSpectrumAnalyzer utility code. The code is currently not working, so I am checking it in disabled, so that John can look at it. I think its just a data type mismatch of some sort.

To enable, set the temporary global variable g_DisableSpectrumMeters to FALSE at the top of NvTerminator.cpp. You can do it before you compile or directly in the debugger after the driver is built. It is very handy for checking timing with and without the functionaly without having to rebuild the driver.

I added the code to the LevelMeter class as we return LM and SM data both at the same time always. There is only 2 function calls for SpectrumMeters (SMs), one for initialization (shared with LevelMeters) and a routine called SpectrumMeter that does all of the SM calculations (currently only 10 band with the other 10 being duplicated at the end). The code looks quite a bit different that the original utility, but functionality should be about the same. Did some name changes, and used our driver variable names i.e. NVA_..., merged a bunch of routines, simplified some code etc...

Note: Even though it is not working yet I ran a quick timing test using VTune. WITHOUT SMs, the nvmcp used 5% of the process time while running DSPlay8 and NvMixer both. of the 5%, the SM routine was only executing 0.3% of that 5%. WITH SMs enabled, nvmcp went up to 11.5% (a full 6.5% higher, more than the entire rest of the driver with VAPU combined). Of the 11.5%, 30% of that time was in the SM rotuine. Obviously this routine needs bookoo optimization before we can use it. No optimizations have been done yet until and wont until it is working.

Change 735255 by mpereira@PEGASUS on 2003/05/29 15:10:34

Summary: added code to support the realtek and empia flexjacks

Change 735235 by speterson@spetersondev on 2003/05/29 15:00:44

Summary: Added 3D volume to the reverb volume so that voice reverbs attenuate properly

Change 734676 by jnorris@jnorris-nf420 on 2003/05/29 10:23:18

Summary: Added the dependent paths for the release build

Change 734512 by gklimanis@gklimanis-driverpull on 2003/05/29 06:16:54

Improved reverb matching with Creative Labs Audigy 2.

Change 734474 by gklimanis@gklimanis-driverpull on 2003/05/29 01:13:57

Reverb matching effort:

1) ReverbCompute_In1Out2.asm :
a) Cleared out remnant output buffers, which contributed to repeating
output of previous frame in combination with output of current frame.
b) Reduced # input diffusors
c) Rearranged feedback of second reverb lung
d) Moved decay filters closer to start of reverb
e) Fixed last delay, which had been excluded, of 2nd lung
f) Restored AddInput code, which is needed for debugging
g) Added right shift to GPSumScaleXXY to reduce clipping

2) Reverb_Util.cpp
a) Added some unparametrized tweaks to shape reverb for longer
decay times. This helps to match Creative Labs Audigy2
reverb behavior for decay times > 10 second
b) Improved sonic quality of reverb -> reduced bounciness
c) Restored reverb decay coefficient to X**(1/4) from x**(1/2)
now that blowup issue doesn't seem to be a problem
d) Added tweak to decay filter for better matching with Creative
labs reverb

3) Util.cpp: forced reverb decay filter to unity gain.

Change 734163 by jweintraub@jweintraub-home on 2003/05/28 18:07:23

VAPU-

1.) Fixed the problems with KMixer streams not playing. With the new way streams are handled, we were precaching (preloading) the ACI buffer up with about 37ms of data (7 interrupt periods). Kmixer streams never have that much data available (where as DirectSound streams have way-way more than that), usually about only 20ms or less. So before a stream is started, if the SSLs do not contain enough data to precache the voice, then the voice is not precached. So for KMixer streams (and any stream with small SSLs) only one ACI buffer worth of data is preloaded (the old way). To do this took yet more logic to control how and when the ACI buffer gets written, and how and when the SSLs are released back to the driver above. There is now (still) possibilities of noise in KMixer streams when interrupts are skipped, as we cannot precache data that does not exist, if anyone has suggestions please let me know.
2.) Added a forth voice list, the new list contains voices that are not available for precaching, which is usually just 2D Kmixer streams.
3.) Fixed some more debug code, to support yet more processing methods.
4.) Some general and miscellaneous cleanup.

HCT testing: current status with this checkin
WaveDrv Tests: 20 Pass, 1 Fail (test 9.42)
DirectSound Tests: 89 Pass, 0 Fail (nice)
KsPosAudio Tests: 8 Pass, 1 Fail (test 1.2, runs when ran standalone)

Change 734146 by jnorris@jnorris-nf420 on 2003/05/28 17:58:48

Summary: Added comb filters to support the VAPU Comb1 and Comb2 preset SoundStorm EQs.

The comb filters use a 4-biquad all-pass filter to create the required logarithmic spacing. Also included is a decorrelator that uses this all-pass filter but this still needs to be plugged in.

All the of the SoundStorm preset EQs are now supported in the VAPU.

Change 734000 by mpereira@PEGASUS on 2003/05/28 16:42:42

Summary: Updated to mute the playback line and mic inputs when those jacks are being used as outputs. This prevents the output of the rear and ctr/lfe channels from getting mixed into the front ouput when they are enabled on the playback mixer. Added code to share the record mute bit with the DRM code. Mute the record line and mic inputs when those jacks are being used as outputs. This prevents recording the surround and ctr/lfe signals when they are selected as the record inputs. This also prevents Digitized input from working on those lines when they are switched to outputs.

Bugid: none

Change 732715 by mpereira@PEGASUS on 2003/05/27 19:22:45

Summary: Update to support jack multiplexing on the Empia (as well as SigmaTel) 6 ch codecs.

Added detection and set a codecoptions flag.
Loaded default register value into reg cache.
Added MXR_CTRL definitions for the codec bits.
Set the startup defaults and handle mixer control requests.

NOTE: The codecs we have to not have real codec VIDs. Also, I am rerouting the MIC line to MIC2 so it doesnt feedback into the main output when swapped.

Bugid: none

Change 732572 by sholmes@wilbur on 2003/05/27 17:53:15

Summary:
Changed the default Kmixer behavior to 48000, 16-bit, stereo with replication (if enabled). This will allow system sounds to get replicated. 6-channel files will be down mixed by the kernel mixer (which uses a crappy down mix making the rears and center very difficult to hear).

Change 732213 by mpereira@PEGASUS on 2003/05/27 15:31:38

Summary: Fix for hang from S4 when SPDIF IN is selected as recording input. Updated to cancel SPDIF IN lock timer and disable SPDIF IN monitor, record, and support when going to nonzero D states.

Bugig: 82927

Change 732084 by sholmes@wilbur on 2003/05/27 14:42:06

BUGID: 82912

Summary:
Fix to increase performance by limiting the kmixer stream so that it doesn't upmix to our best format but don't have it enable (for reason to be explained). Also made it so the kmixer stream would get replicated to the rears when the fix is enabled. This required some addition bug fixes to nvvoice.cpp since routing wasn't correct for replicated stereo voices.

Fix is not enabled becuase when it is, kmixer down mixes 6-channel to 2-channel incorrectly and I can not determine why. It is placing the back left into the right channel for some reason and I've verifed this by recording the data being passed to the driver.

Change 732037 by speterson@spetersondev on 2003/05/27 14:22:50

Summary: Reenabled logarithmic scaling of volume sliders for NT4.0
BugID: 83053

Change 731949 by speterson@spetersondev on 2003/05/27 13:52:38

Summary: Adding inf files for MegaPC version of SOundStorm control panel with AUX labled as Radio

Change 731925 by speterson@spetersondev on 2003/05/27 13:42:35

Summary: Added guids for AUX mixer node so that it's name can be altered through the inf file.
Changed SPDIF input record gain slider to be non-functional since it has no effect on volume.

Change 731754 by builder@BUILDERNV4_8 on 2003/05/27 11:38:15

Updated version to 3.43

Change 731735 by builder@BUILDERNV4_8 on 2003/05/27 11:26:21

Updated DriverVer to 05/27/2003