nouveau Wiki - jwstolk

Jaap Stolk

2010 Update

The hardware below (including the NV15) card stopped working.
The links to http://jwstolk.xs4all.nl/ are also down, but I can e-mail Octave plots used to find the spotlight parameters.
I included the C-code for calculating the 7 hardware registers from the GL parameters below. Measurements produced a strange function, and showed many non-linearities, which could very well be the result of using a look-up-table and using linear interpolation on that table. We worked out the relations between the parameters, so only two tables where needed to calculate the 7 hardware registers. The table values where trained using a tail-and-error tool, until they fitted the measurements very well. (Note that some global variable could affect the needed circulations, we only tried (many) combinations of spotlight parameters.)

Hardware

AMD Athlon 1200MHz, 640MB, Gentoo, kernel 2.6.17.6
nv15 GeForce2 GTS/Pro rev a3

lspci:

01:00.0 VGA compatible controller: nVidia Corporation NV15 [!GeForce2 GTS/Pro] (rev a3) (prog-if 00 [VGA])
        Subsystem: ASUSTeK Computer Inc. Unknown device 400e
        Flags: bus master, 66MHz, medium devsel, latency 32, IRQ 10
        Memory at dc000000 (32-bit, non-prefetchable) [size=16M]
        Memory at d0000000 (32-bit, prefetchable) [size=128M]
        [virtual] Expansion ROM at dd000000 [disabled] [size=64K]
        Capabilities: [60] Power Management version 1
        Capabilities: [44] AGP version 2.0

spotlight parameters

plots used for decoding the spotlight parameters: http://jwstolk.xs4all.nl/nouveau/plots.htm
and: http://jwstolk.xs4all.nl/nouveau/plots_old.htm
latest spotlight code: http://jwstolk.xs4all.nl/nouveau/code/spotlight_parameters.c
other code (used to test and adjust the look-Up-Tables) code: http://jwstolk.xs4all.nl/nouveau/code/

mmio-trace file format converter:

http://jwstolk.xs4all.nl/nouveau/code/mmio-convert.c

mmio-trace macro search program:

many ideas, not a lot of code.

Testing in-kernel-tree mmio-trace module with kernel 2.6.24 and Nouveau:

(Note: Tracing nouveau is only academic, we could just look at the source. Maybe useful to compare with a real trace.)
Get 2.6.24 kernel (Gentoo: git-sources)
Apply pq's page fault hook patch and mmio-trace module patch to the kernel.
Enable DebugFS (CONFIG_DEBUG_FS), RelayFS (CONFIG_RELAY), CONFIG_PAGE_FAULT_HANDLERS and CONFIG_MMIOTRACE.
Compile and install kernel.
Follow InstallNouveau but don't insmod anything and don't startx. (use the ./configure --prefix=/usr/ hint for drm on Gentoo)
Get latest mmio-trace: git clone git://people.freedesktop.org/~pq/mmio-trace
Build mmio-trace: cd mmio-trace; make userspace (we use the in-kernel-tree module instead)
Make a hooked copy of the drm and nouveau module:
- ./hook-module <...>/drm/linux-core/drm.ko drm-hook.ko
- ./hook-module <...>/drm/linux-core/nouveau.ko nouveau-hook.ko
Mount debugfs to /debug: mkdir -p /debug; mount -t debugfs debugfs /debug
Load the mmio module: insmod /lib/modules/2.6.24-rc8-git7/kernel/arch/x86/kernel/mmiotrace/mmiotrace.ko
Start ./mmio-trace, it will create file cpu0 on the current directory.
Load the hooked nvidia module: insmod drm-hook.ko insmod nouveau-hook.ko
Start Xorg, do some simple things for a few seconds, and quit Xorg. Now you have a trace, and all that is left is to shut down tracing.
Unload the hooked modules: rmmod nouveau; rmmod drm
Kill the mmio-trace process. Ctrl-c, for instance.
Unload mmio module: rmmod mmio
Get the lxml package (if needed)
Get the rules-ng register name database: cvs -z3 -d:pserver:anonymous@nouveau.cvs.sourceforge.net:/cvsroot/nouveau co -P rules-ng
(fix: chmod +x rules-ng/parsers/*.py)
Compile the staic databse as a lib: cd rules-ng/staticdb; make; cp libnvidia-mmio.so <...>/mmio-trace
./mmio-parse -m NN -s ./libnvidia-mmio.so < cpu0 > cpu0_parsed (NN = 15 for a NV15 card, etc.)

irc: freenode, #nouveau and #mmio-trace, nick: jwstolk

mail: jwstolk_at_gmail.

//  spotlight_parameters.c
//
//  Copyright (C) 2007 J.W. Stolk, jwstolk@gmail.com, GPL v2
//
//  -----------------------------------------------------------------------
//  Description: spotlight functions for the Nouveau 3D driver
//  see: http://nouveau.cvs.sourceforge.net/nouveau/doc/nv20_light
//       http://jwstolk.xs4all.nl/nouveau/?M=A
//       http://jwstolk.xs4all.nl/nouveau/plots.htm
//       http://jwstolk.xs4all.nl/nouveau/plots_old.htm
//       http://jwstolk.xs4all.nl/nouveau/code/
//  -----------------------------------------------------------------------
//  compile: gcc spotlight_parameters.c -lm -Wall -ospotlight_parameters
//  -----------------------------------------------------------------------
//  version: Sun Feb  4 21:24:50 CET 2007

#include <math.h>
#include <stdio.h>

// ,----------------------------------------------------------------------,
// | Input:  GL parameters:      exponent, cutoff, dirX, dirY, dirZ       |
// | Output: hardware registers: hw_reg_a, hw_reg_b, hw_reg_c,            |
// |                             hw_reg_d, hw_reg_e, hw_reg_f, hw_reg_g   |
// '----------------------------------------------------------------------'

void get_spotlight_parameters(double exponent, double cutoff, double dirX, double dirY, double dirZ,
                              double* hw_reg_a, double* hw_reg_b, double* hw_reg_c,
                              double* hw_reg_d, double* hw_reg_e, double* hw_reg_f, double* hw_reg_g){

  // The hardware registers are 32-bit floating point, so using double is more than sufficent.
  double cutoff_rad;
  double lut_index;     // (renamed from step_nr)
  int    lut_index_int;
  double lut_value;
  double left_factor;
  double right_factor;
  double spotlight_length;
  double tmp;

  // Look-Up-Tables:
const double spotlight_LUT[51]={
  -0.00004313671710,-0.00001664488452,-0.00001692090548,-0.00001692090548,-0.00001692090548,
  -0.00001692090548,-0.00001692090547,-0.00475650768720,-0.00949630692880,-0.01423604510371,
  -0.01897580409635,-0.02371554958185,-0.02995794597311,-0.03521003198499,-0.03967819537527,
  -0.04331929838488,-0.04649309566914,-0.04914397523183,-0.05136181729088,-0.05328106166245,
  -0.05492743778197,-0.05612668233260,-0.05696293343000,-0.05703870874157,-0.05605074400074,
  -0.05367270272221,-0.04924597960194,-0.04221648841730,-0.03196529891062,-0.01801446831990,
  -0.00001176230050, 0.03771627483479, 0.09906659099714, 0.17749723974919, 0.26660459589762,
   0.36026175679746, 0.45244459111994, 0.53970871355655, 0.61833349227586, 0.68679650633229,
   0.74600077835005, 0.79561688860929, 0.83640888233911, 0.87019888840919, 0.89667101785942,
   0.91862913345432, 0.93576909519174, 0.94930296515212, 0.96040906843445, 0.96858874831011,
   0.97572753616790};
const double exponent_LUT[51] ={
   0.99999434105398, 0.99999302624596, 0.99999606609345, 0.99999606609344, 0.99999606609345,
   0.99999606609344, 0.99999606609345, 0.99961048683387, 0.99922497235411, 0.99883938661131,
   0.99845380095804, 0.99806823281838, 0.99708892684478, 0.99581822547003, 0.99414524850779,
   0.99195363947614, 0.98902742170359, 0.98511916671572, 0.97986437860412, 0.97270389416853,
   0.96285086481830, 0.94925040677666, 0.93017427428989, 0.90341705531002, 0.86563085422000,
   0.81217929697770, 0.73670241935000, 0.63063384884101, 0.48277542009796, 0.27857488415982,
  -0.00003948562476,-0.31875989523469,-0.64839014406671,-0.98119795202123,-1.30545553528001,
  -1.60888090900001,-1.88661310700404,-2.12615570648801,-2.33539034461990,-2.52149001443213,
  -2.66658392901470,-2.78644657501522,-2.89098264098132,-2.96975242355677,-3.04852169688304,
  -3.08190421360393,-3.13682821311390,-3.19654141784206,-3.24449362878119,-3.29244580044634,
  -3.35922785872215};

const double exp_conv_factor = 0.2482711787;
  
  // calculate the index for the Look-Up-Table:
  // note: inverse of: exponent = exp((lut_index-30)*exp_conv_factor);
  lut_index = log(exponent)/exp_conv_factor + 30;
  lut_index_int = floor(lut_index);
  right_factor = (double) lut_index - lut_index_int;
  left_factor  = 1.0 - right_factor;

  // clamp exponent to range of Look-Up_Table.  valid exponent range:  0.0005825 ... 143.369264
  if( lut_index_int < 0 ){
    lut_index    = 0;
    lut_index_int= 0;
    right_factor = 0;
    left_factor  = 1;
  }else if( lut_index_int > 49 ){
    // (should not be possible. max exponent=128)
    lut_index    = 50;
    lut_index_int= 49;
    right_factor = 1;
    left_factor  = 0;
  }

  // hw_reg_b is a direct lookup based on exponent, it's not affected by other parameters:
  *hw_reg_b = exponent_LUT[lut_index_int]*left_factor + exponent_LUT[lut_index_int+1]*right_factor;

  // lookup the initial hw_reg_a and hw_reg_g, based on exponent:
  lut_value = spotlight_LUT[lut_index_int]*left_factor + spotlight_LUT[lut_index_int+1]*right_factor;

  // adjust for cutoff:
  cutoff_rad = cutoff*M_PI/180.0; 
  *hw_reg_a = (cos(cutoff_rad)-lut_value) / (1-cos(cutoff_rad));

  // all hw_reg's (except hw_reg_b) change at the hw_reg_a == 0 point:
  if( *hw_reg_a<0.0 ){
    *hw_reg_a = 0;
  }else{
    // if hw_reg_a > 0, then hw_reg_g should be calculated for the exponent at hw_reg_a==0
    // solve: w_reg_a == (cosl(cutoff_rad)-lut_value) / (1-cosl(cutoff_rad)) == 0
    lut_value = cos(cutoff_rad);
    // i = 23;   this is the lowest point in the LUT. start search from here;
    // while( spotlight_LUT[i] < lut_value_a0 && i<50 ) i++;
    // reverse-interpolate:
    // lut_index_a0 = i-1 + (lut_value_a0-spotlight_LUT[i-1]) / (spotlight_LUT[i-1] - spotlight_LUT[i]);
    // exponent_a0 = exp( (lut_index-30)*exp_conv_factor );
    // but we don't need the exponent, because hw_reg_g uses the same look-up-table
    // so lut_value can directly be used to calculate hw_reg_g
  }

  // hw_reg_c always is: hw_reg_a - hw_reg_b + 1:
  *hw_reg_c = *hw_reg_a - *hw_reg_b + 1.0;

  // hw_reg_g:
  tmp = 1/(lut_value-1);
  *hw_reg_g = tmp + 1;     // hw_reg_g = 1/(lut_value-1) + 1;

  // hw_reg_d, hw_reg_e, hw_reg_f:
  spotlight_length = sqrt( dirX*dirX + dirY*dirY + dirZ*dirZ );
  tmp /= spotlight_length;
  *hw_reg_d = tmp * dirX;  // hw_reg_d = 1/(lut_value-1) * dirX/spotlight_length;
  *hw_reg_e = tmp * dirY;  // hw_reg_e = 1/(lut_value-1) * dirY/spotlight_length;
  *hw_reg_f = tmp * dirZ;  // hw_reg_f = 1/(lut_value-1) * dirZ/spotlight_length;

}

// ,----------------------------------------------------------------------,
// |          Main:                                                       |
// '----------------------------------------------------------------------'

int main(int argc, char **argv){

  double exponent;
  double cutoff;
  double dirX;
  double dirY;
  double dirZ;
  double hw_reg_a;
  double hw_reg_b;
  double hw_reg_c;
  double hw_reg_d;
  double hw_reg_e;
  double hw_reg_f;
  double hw_reg_g;

  exponent = 1.52;
  cutoff = 20;
  dirX = 0.50;
  dirY = 0.51;
  dirZ = 0.52;

  get_spotlight_parameters(exponent, cutoff, dirX, dirY, dirZ, &hw_reg_a, &hw_reg_b, &hw_reg_c,
                           &hw_reg_d, &hw_reg_e, &hw_reg_f, &hw_reg_g);

  printf("hw_reg_a = %15.10f\n",hw_reg_a);
  printf("hw_reg_b = %15.10f\n",hw_reg_b);
  printf("hw_reg_c = %15.10f\n",hw_reg_c);
  printf("hw_reg_d = %15.10f\n",hw_reg_d);
  printf("hw_reg_e = %15.10f\n",hw_reg_e);
  printf("hw_reg_f = %15.10f\n",hw_reg_f);
  printf("hw_reg_g = %15.10f\n",hw_reg_g);

  return(0);
}

CategoryHomepage

jwstolk

Jaap Stolk

2010 Update

Hardware

Links

spotlight parameters

mmio-trace file format converter:

mmio-trace macro search program:

Testing in-kernel-tree mmio-trace module with kernel 2.6.24 and Nouveau: