Merge remote branch 'upstream/gallium-0.1' into nouveau-gallium-0.1

16 files changed, 306 insertions, 68 deletions

diff --git a/progs/demos/texcyl.c b/progs/demos/texcyl.c
index c04d5004e37..e3b03039759 100644
--- a/progs/demos/texcyl.c
+++ b/progs/demos/texcyl.c
@@ -27,6 +27,8 @@
 #define LINEAR_FILTER 21
 #define QUIT 100
 
+static GLint Win = -1;
+
 static GLuint CylinderObj = 0;
 static GLboolean Animate = GL_TRUE;
 
@@ -159,6 +161,7 @@ static void Key( unsigned char key, int x, int y )
          glutIdleFunc(NULL);
       break;
       case 27:
+         glutDestroyWindow(Win);
          exit(0);
          break;
    }
@@ -263,7 +266,7 @@ int main( int argc, char *argv[] )
 
    glutInitDisplayMode( GLUT_RGB | GLUT_DOUBLE );
 
-   glutCreateWindow(argv[0] );
+   Win = glutCreateWindow(argv[0] );
 
    Init(argc, argv);
 
diff --git a/progs/tests/Makefile b/progs/tests/Makefile
index ea34a708553..27e2eafdfcb 100644
--- a/progs/tests/Makefile
+++ b/progs/tests/Makefile
@@ -40,6 +40,7 @@ SOURCES = \
 	fbotest1.c \
 	fbotest2.c \
 	fbotexture.c \
+	fillrate.c \
 	fog.c \
 	fogcoord.c \
 	fptest1.c \
@@ -166,6 +167,13 @@ mipmap_view.o: mipmap_view.c readtex.h
 	$(CC) -c $(INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
 
 
+fillrate: fillrate.o readtex.o
+	$(CC) $(CFLAGS) fillrate.o readtex.o $(LIBS) -o $@
+
+fillrate.o: fillrate.c readtex.h
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
+
+
 
 readtex.o: readtex.c
 	$(CC) -c $(INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
diff --git a/progs/tests/fillrate.c b/progs/tests/fillrate.c
new file mode 100644
index 00000000000..8fe636c364b
--- /dev/null
+++ b/progs/tests/fillrate.c
@@ -0,0 +1,203 @@
+/**
+ * Measure fill rates for basic shading/texturing modes.
+ *
+ * Brian Paul
+ * 1 April 2008
+ */
+
+
+#define GL_GLEXT_PROTOTYPES
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <GL/glut.h>
+#include "readtex.h"
+
+#define TEXTURE_1_FILE "../images/tile.rgb"
+#define TEXTURE_2_FILE "../images/reflect.rgb"
+
+static int Win;
+static int Width = 1010, Height = 1010;
+static GLuint Textures[2];
+
+
+/**
+ * Draw quad 10 pixels shorter, narrower than window size.
+ */
+static void
+DrawQuad(void)
+{
+   glBegin(GL_POLYGON);
+
+   glColor3f(1.0, 0.5, 0.5);
+   glMultiTexCoord2f(GL_TEXTURE0, 0, 0);
+   glMultiTexCoord2f(GL_TEXTURE1, 0, 0);
+   glVertex2f(5, 5);
+
+   glColor3f(0.5, 1.0, 0.5);
+   glMultiTexCoord2f(GL_TEXTURE0, 1, 0);
+   glMultiTexCoord2f(GL_TEXTURE1, 1, 0);
+   glVertex2f(Width - 5, 5);
+
+   glColor3f(0.5, 0.5, 1.0);
+   glMultiTexCoord2f(GL_TEXTURE0, 1, 1);
+   glMultiTexCoord2f(GL_TEXTURE1, 1, 1);
+   glVertex2f(Width - 5, Height - 5);
+
+   glColor3f(1.0, 0.5, 1.0);
+   glMultiTexCoord2f(GL_TEXTURE0, 0, 1);
+   glMultiTexCoord2f(GL_TEXTURE1, 0, 1);
+   glVertex2f(5, Height - 5);
+
+   glEnd();
+}
+
+
+/**
+ * Compute rate for drawing large quad with given shading/texture state.
+ */
+static void
+RunTest(GLenum shading, GLuint numTextures, GLenum texFilter)
+{
+   const GLdouble minPeriod = 2.0;
+   GLdouble t0, t1;
+   GLdouble pixels, rate;
+   GLint i, iters;
+
+   glActiveTexture(GL_TEXTURE0);
+   if (numTextures > 0) {
+      glBindTexture(GL_TEXTURE_2D, Textures[0]);
+      glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, texFilter);
+      glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, texFilter);
+      glEnable(GL_TEXTURE_2D);
+   }
+   else {
+      glDisable(GL_TEXTURE_2D);
+   }
+
+   glActiveTexture(GL_TEXTURE1);
+   if (numTextures > 1) {
+      glBindTexture(GL_TEXTURE_2D, Textures[1]);
+      glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, texFilter);
+      glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, texFilter);
+      glEnable(GL_TEXTURE_2D);
+   }
+   else {
+      glDisable(GL_TEXTURE_2D);
+   }
+
+   glShadeModel(shading);
+
+
+   glFinish();
+
+   iters = 1;
+   do {
+      iters *= 4;
+      t0 = glutGet(GLUT_ELAPSED_TIME) * 0.001;
+      for (i = 0; i < iters; i++) {
+         DrawQuad();
+      }
+      glFinish();
+      t1 = glutGet(GLUT_ELAPSED_TIME) * 0.001;
+   } while (t1 - t0 < minPeriod);
+
+   glutSwapBuffers();
+
+   pixels = (double) iters * (Width - 10) * (Height - 10);
+   rate = pixels / (t1 - t0);
+   rate /= 1000000.0;  /* megapixels/second */
+
+   printf("%s ", shading == GL_FLAT ? "GL_FLAT" : "GL_SMOOTH");
+   printf("Textures=%u ", numTextures);
+   printf("Filter=%s: ", texFilter == GL_LINEAR ? "GL_LINEAR" : "GL_NEAREST");
+   printf("%g MPixels/sec\n", rate);
+}
+
+
+static void
+Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT);
+
+   RunTest(GL_FLAT, 0, GL_NEAREST);
+   RunTest(GL_SMOOTH, 0, GL_NEAREST);
+   RunTest(GL_SMOOTH, 1, GL_NEAREST);
+   RunTest(GL_SMOOTH, 1, GL_LINEAR);
+   RunTest(GL_SMOOTH, 2, GL_NEAREST);
+   RunTest(GL_SMOOTH, 2, GL_LINEAR);
+
+   glutSwapBuffers();
+}
+
+
+static void
+Reshape(int width, int height)
+{
+   glViewport(0, 0, width, height);
+   glMatrixMode(GL_PROJECTION);
+   glLoadIdentity();
+   glOrtho(0, width, 0, height, -1, 1);
+   glMatrixMode(GL_MODELVIEW);
+   glLoadIdentity();
+   Width = width;
+   Height = height;
+}
+
+
+static void
+Key(unsigned char key, int x, int y)
+{
+   (void) x;
+   (void) y;
+   switch (key) {
+   case 27:
+      glutDestroyWindow(Win);
+      exit(0);
+      break;
+   }
+   glutPostRedisplay();
+}
+
+
+static void
+Init(void)
+{
+   printf("GL_RENDERER = %s\n", (char *) glGetString(GL_RENDERER));
+
+   glGenTextures(2, Textures);
+
+   glBindTexture(GL_TEXTURE_2D, Textures[0]);
+   glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);
+
+   if (!LoadRGBMipmaps(TEXTURE_1_FILE, GL_RGB)) {
+      printf("Error: couldn't load texture image\n");
+      exit(1);
+   }
+
+   glBindTexture(GL_TEXTURE_2D, Textures[1]);
+   glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_MODULATE);
+
+   if (!LoadRGBMipmaps(TEXTURE_2_FILE, GL_RGB)) {
+      printf("Error: couldn't load texture image\n");
+      exit(1);
+   }
+
+}
+
+
+int
+main(int argc, char *argv[])
+{
+   glutInit(&argc, argv);
+   glutInitWindowPosition(0, 0);
+   glutInitWindowSize(Width, Height);
+   glutInitDisplayMode(GLUT_RGB | GLUT_DOUBLE);
+   Win = glutCreateWindow(argv[0]);
+   glutReshapeFunc(Reshape);
+   glutKeyboardFunc(Key);
+   glutDisplayFunc(Draw);
+   Init();
+   glutMainLoop();
+   return 0;
+}
diff --git a/src/gallium/auxiliary/draw/draw_prim.c b/src/gallium/auxiliary/draw/draw_prim.c
index 4452376a70e..51b69503344 100644
--- a/src/gallium/auxiliary/draw/draw_prim.c
+++ b/src/gallium/auxiliary/draw/draw_prim.c
@@ -119,43 +119,6 @@ static void draw_prim_queue_flush( struct draw_context *draw )
    draw_vertex_cache_unreference( draw );
 }
 
-static INLINE void fetch_and_store(struct draw_context *draw)
-{
-   /* run vertex shader on vertex cache entries, four per invokation */
-#if 0
-   {
-      const struct vertex_info *vinfo = draw->render->get_vertex_info(draw->render);
-      memcpy(draw->vs.queue[0].vertex, draw->vs.queue[i + j].elt,
-             count * vinfo->size);
-   }
-#elif 0
-   unsigned i;
-
-   draw_update_vertex_fetch(draw);
-   for (i = 0; i < draw->vs.queue_nr; i += 4) {
-      struct vertex_header *dests[4];
-      unsigned elts[4];
-      struct tgsi_exec_machine *machine = &draw->machine;
-      int j, n = MIN2(4, draw->vs.queue_nr - i);
-
-      for (j = 0; j < n; j++) {
-         elts[j] = draw->vs.queue[i + j].elt;
-         dests[j] = draw->vs.queue[i + j].vertex;
-      }
-
-      for ( ; j < 4; j++) {
-	 elts[j] = elts[0];
-         dests[j] = draw->vs.queue[i + j].vertex;
-      }
-      //fetch directly into dests
-      draw->vertex_fetch.fetch_func(draw, machine, dests, count);
-   }
-#endif
-
-   draw->vs.post_nr = draw->vs.queue_nr;
-   draw->vs.queue_nr = 0;
-}
-
 void draw_do_flush( struct draw_context *draw, unsigned flags )
 {
    if (0)
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index 52c31260505..67b87f16d7f 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -232,12 +232,12 @@ cell_bind_sampler_states(struct pipe_context *pipe,
 {
    struct cell_context *cell = cell_context(pipe);
 
-   draw_flush(cell->draw);
+   assert(num <= CELL_MAX_SAMPLERS);
 
-   assert(unit < PIPE_MAX_SAMPLERS);
+   draw_flush(cell->draw);
 
    memcpy(cell->sampler, samplers, num * sizeof(void *));
-   memset(&cell->sampler[num], 0, (PIPE_MAX_SAMPLERS - num) *
+   memset(&cell->sampler[num], 0, (CELL_MAX_SAMPLERS - num) *
           sizeof(void *));
    cell->num_samplers = num;
 
@@ -261,6 +261,8 @@ cell_set_sampler_textures(struct pipe_context *pipe,
    struct cell_context *cell = cell_context(pipe);
    uint i;
 
+   assert(num <= CELL_MAX_SAMPLERS);
+
    /* Check for no-op */
    if (num == cell->num_textures &&
        !memcmp(cell->texture, texture, num * sizeof(struct pipe_texture *)))
@@ -268,7 +270,7 @@ cell_set_sampler_textures(struct pipe_context *pipe,
 
    draw_flush(cell->draw);
 
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+   for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
       struct pipe_texture *tex = i < num ? texture[i] : NULL;
 
       pipe_texture_reference((struct pipe_texture **) &cell->texture[i], tex);
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c
index 84b48bf4f1b..5198b51441a 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.c
+++ b/src/gallium/drivers/cell/ppu/cell_screen.c
@@ -31,6 +31,7 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
 
+#include "cell/common.h"
 #include "cell_screen.h"
 #include "cell_texture.h"
 #include "cell_winsys.h"
@@ -55,7 +56,7 @@ cell_get_param(struct pipe_screen *screen, int param)
 {
    switch (param) {
    case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
-      return PIPE_MAX_SAMPLERS;
+      return CELL_MAX_SAMPLERS;
    case PIPE_CAP_NPOT_TEXTURES:
       return 0;
    case PIPE_CAP_TWO_SIDED_STENCIL:
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index 9c694e136d8..07717da8f6a 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -186,15 +186,17 @@ tile_copy_data(uint w, uint h, uint tile_size, uint *dst, const uint *src)
    uint it, jt;  /* tile counters */
    uint i, j;    /* intra-tile counters */
 
+   /* loop over dest tiles */
    for (it = 0; it < h_t; it++) {
       for (jt = 0; jt < w_t; jt++) {
-         /* fill in tile (i, j) */
+         /* start of dest tile: */
          uint *tdst = dst + (it * w_t + jt) * tile_size2;
+         /* loop over texels in the tile */
          for (i = 0; i < tile_size; i++) {
             for (j = 0; j < tile_size; j++) {
                const uint srci = it * tile_size + i;
                const uint srcj = jt * tile_size + j;
-               *tdst++ = src[srci * h + srcj];
+               *tdst++ = src[srci * w + srcj];
             }
          }
       }
@@ -245,9 +247,13 @@ void
 cell_update_texture_mapping(struct cell_context *cell)
 {
    uint face = 0, level = 0, zslice = 0;
+   uint i;
+
+   for (i = 0; i < CELL_MAX_SAMPLERS; i++) {
+      if (cell->texture[i])
+         cell_tile_texture(cell, cell->texture[i]);
+   }
 
-   if (cell->texture[0])
-      cell_tile_texture(cell, cell->texture[0]);
 #if 0
    if (cell->tex_surf && cell->tex_map) {
       pipe_surface_unmap(cell->tex_surf);
diff --git a/src/gallium/drivers/cell/spu/spu_dcache.c b/src/gallium/drivers/cell/spu/spu_dcache.c
index a1701d80d18..167404cdc54 100644
--- a/src/gallium/drivers/cell/spu/spu_dcache.c
+++ b/src/gallium/drivers/cell/spu/spu_dcache.c
@@ -36,6 +36,7 @@
 #define CACHE_SET_TAGID(set)  (((set) & 0x03) + TAG_DCACHE0)
 #define CACHE_LOG2NNWAY       2
 #define CACHE_LOG2NSETS       6
+/*#define CACHE_STATS           1*/
 #include <cache-api.h>
 
 /* Yes folks, this is ugly.
@@ -123,3 +124,20 @@ spu_dcache_mark_dirty(unsigned ea, unsigned size)
           ? (entry & ~CACHELINE_VALID) : entry;
    }
 }
+
+
+/**
+ * Print cache utilization report
+ */
+void
+spu_dcache_report(void)
+{
+#ifdef CACHE_STATS
+   if (spu.init.id == 0) {
+      printf("SPU 0: Texture cache report:\n");
+      cache_pr_stats(data);
+   }
+#endif
+}
+
+
diff --git a/src/gallium/drivers/cell/spu/spu_dcache.h b/src/gallium/drivers/cell/spu/spu_dcache.h
index 7a06b8c25af..39a19eb31b5 100644
--- a/src/gallium/drivers/cell/spu/spu_dcache.h
+++ b/src/gallium/drivers/cell/spu/spu_dcache.h
@@ -31,4 +31,7 @@ spu_dcache_fetch_unaligned(qword *dst, unsigned ea, unsigned size);
 extern void
 spu_dcache_mark_dirty(unsigned ea, unsigned size);
 
+extern void
+spu_dcache_report(void);
+
 #endif /* SPU_DCACHE_H */
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 7f0473d198c..1ab1c40379a 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -319,10 +319,10 @@ cmd_state_sampler(const struct cell_command_sampler *sampler)
              spu.init.id, sampler->unit);
 
    spu.sampler[sampler->unit] = sampler->state;
-   if (spu.sampler[0].min_img_filter == PIPE_TEX_FILTER_LINEAR)
-      spu.sample_texture = sample_texture_bilinear;
+   if (spu.sampler[sampler->unit].min_img_filter == PIPE_TEX_FILTER_LINEAR)
+      spu.sample_texture[sampler->unit] = sample_texture_bilinear;
    else
-      spu.sample_texture = sample_texture_nearest;
+      spu.sample_texture[sampler->unit] = sample_texture_nearest;
 }
 
 
@@ -629,6 +629,8 @@ main_loop(void)
 
    if (Debug)
       printf("SPU %u: Exit main loop\n", spu.init.id);
+
+   spu_dcache_report();
 }
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 2bfad3535a1..e9e39cbeab3 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -85,6 +85,8 @@ typedef struct spu_blend_results (*logicop_func)(
     qword frag_mask);
 
 
+typedef vector float (*sample_texture_func)(uint unit, vector float texcoord);
+
 struct spu_framebuffer {
    void *color_start;              /**< addr of color surface in main memory */
    void *depth_start;              /**< addr of depth surface in main memory */
@@ -152,7 +154,7 @@ struct spu_global
    /** for converting RGBA to PIPE_FORMAT_x colors */
    vector unsigned char color_shuffle;
 
-   vector float (*sample_texture)(vector float texcoord);
+   sample_texture_func sample_texture[CELL_MAX_SAMPLERS];
 
 } ALIGN16_ATTRIB;
 
diff --git a/src/gallium/drivers/cell/spu/spu_texture.c b/src/gallium/drivers/cell/spu/spu_texture.c
index 4612501eb3b..ceab2469803 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.c
+++ b/src/gallium/drivers/cell/spu/spu_texture.c
@@ -49,9 +49,8 @@ invalidate_tex_cache(void)
 
 
 static uint
-get_texel(vec_uint4 coordinate)
+get_texel(uint unit, vec_uint4 coordinate)
 {
-   const uint unit = 0;
    vec_uint4 tmp;
    unsigned x = spu_extract(coordinate, 0);
    unsigned y = spu_extract(coordinate, 1);
@@ -69,9 +68,8 @@ get_texel(vec_uint4 coordinate)
 
 
 static void
-get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
+get_four_texels(uint unit, vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
 {
-   const uint unit = 0;
    const unsigned texture_ea = (uintptr_t) spu.texture[unit].start;
    vec_uint4 tile_x = spu_rlmask(x, -5);
    vec_uint4 tile_y = spu_rlmask(y, -5);
@@ -104,21 +102,19 @@ get_four_texels(vec_uint4 x, vec_uint4 y, vec_uint4 *texels)
  * XXX this is extremely primitive for now.
  */
 vector float
-sample_texture_nearest(vector float texcoord)
+sample_texture_nearest(uint unit, vector float texcoord)
 {
-   const uint unit = 0;
    vector float tc = spu_mul(texcoord, spu.texture[unit].tex_size);
    vector unsigned int itc = spu_convtu(tc, 0);  /* convert to int */
    itc = spu_and(itc, spu.texture[unit].tex_size_mask); /* mask (GL_REPEAT) */
-   uint texel = get_texel(itc);
+   uint texel = get_texel(unit, itc);
    return spu_unpack_A8R8G8B8(texel);
 }
 
 
 vector float
-sample_texture_bilinear(vector float texcoord)
+sample_texture_bilinear(uint unit, vector float texcoord)
 {
-   const uint unit = 0;
    static const vec_uint4 offset_x = {0, 0, 1, 1};
    static const vec_uint4 offset_y = {0, 1, 0, 1};
 
@@ -139,7 +135,7 @@ sample_texture_bilinear(vector float texcoord)
    x = spu_and(x, spu.texture[unit].tex_size_x_mask);
    y = spu_and(y, spu.texture[unit].tex_size_y_mask);
 
-   get_four_texels(x, y, texels);
+   get_four_texels(unit, x, y, texels);
 
    vector float texel00 = spu_unpack_A8R8G8B8(spu_extract(texels[0], 0));
    vector float texel01 = spu_unpack_A8R8G8B8(spu_extract(texels[1], 0));
diff --git a/src/gallium/drivers/cell/spu/spu_texture.h b/src/gallium/drivers/cell/spu/spu_texture.h
index 95eb87080f1..f7c9738be88 100644
--- a/src/gallium/drivers/cell/spu/spu_texture.h
+++ b/src/gallium/drivers/cell/spu/spu_texture.h
@@ -37,11 +37,11 @@ invalidate_tex_cache(void);
 
 
 extern vector float
-sample_texture_nearest(vector float texcoord);
+sample_texture_nearest(uint unit, vector float texcoord);
 
 
 extern vector float
-sample_texture_bilinear(vector float texcoord);
+sample_texture_bilinear(uint unit, vector float texcoord);
 
 
 #endif /* SPU_TEXTURE_H */
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 17e337bbdf6..ab4ff8160a9 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -311,17 +311,43 @@ emit_quad( int x, int y, mask_t mask )
 
       if (spu.texture[0].start) {
          /* texture mapping */
+         const uint unit = 0;
          vector float texcoords[4];
          eval_coeff(2, (float) x, (float) y, texcoords);
 
          if (spu_extract(mask, 0))
-            colors[0] = spu.sample_texture(texcoords[0]);
+            colors[0] = spu.sample_texture[unit](unit, texcoords[0]);
          if (spu_extract(mask, 1))
-            colors[1] = spu.sample_texture(texcoords[1]);
+            colors[1] = spu.sample_texture[unit](unit, texcoords[1]);
          if (spu_extract(mask, 2))
-            colors[2] = spu.sample_texture(texcoords[2]);
+            colors[2] = spu.sample_texture[unit](unit, texcoords[2]);
          if (spu_extract(mask, 3))
-            colors[3] = spu.sample_texture(texcoords[3]);
+            colors[3] = spu.sample_texture[unit](unit, texcoords[3]);
+
+
+         if (spu.texture[1].start) {
+            /* multi-texture mapping */
+            const uint unit = 1;
+            vector float colors1[4];
+
+            eval_coeff(2, (float) x, (float) y, texcoords);
+
+            if (spu_extract(mask, 0))
+               colors1[0] = spu.sample_texture[unit](unit, texcoords[0]);
+            if (spu_extract(mask, 1))
+               colors1[1] = spu.sample_texture[unit](unit, texcoords[1]);
+            if (spu_extract(mask, 2))
+               colors1[2] = spu.sample_texture[unit](unit, texcoords[2]);
+            if (spu_extract(mask, 3))
+               colors1[3] = spu.sample_texture[unit](unit, texcoords[3]);
+
+            /* hack: modulate first texture by second */
+            colors[0] = spu_mul(colors[0], colors1[0]);
+            colors[1] = spu_mul(colors[1], colors1[1]);
+            colors[2] = spu_mul(colors[2], colors1[2]);
+            colors[3] = spu_mul(colors[3], colors1[3]);
+         }
+
       }
       else {
          /* simple shading */
diff --git a/src/mesa/state_tracker/st_extensions.c b/src/mesa/state_tracker/st_extensions.c
index 7fab4d041d0..20f4652413d 100644
--- a/src/mesa/state_tracker/st_extensions.c
+++ b/src/mesa/state_tracker/st_extensions.c
@@ -83,7 +83,8 @@ void st_init_limits(struct st_context *st)
    c->MaxTextureRectSize
       = min(1 << (c->MaxTextureLevels - 1), MAX_TEXTURE_RECT_SIZE);
 
-   c->MaxTextureImageUnits
+   c->MaxTextureUnits
+      = c->MaxTextureImageUnits
       = c->MaxTextureCoordUnits
       = min(screen->get_param(screen, PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS),
             MAX_TEXTURE_IMAGE_UNITS);
diff --git a/src/mesa/state_tracker/st_program.c b/src/mesa/state_tracker/st_program.c
index 101a6a48afd..4d1ef1bf3d3 100644
--- a/src/mesa/state_tracker/st_program.c
+++ b/src/mesa/state_tracker/st_program.c
@@ -34,6 +34,7 @@
 #include "main/imports.h"
 #include "main/mtypes.h"
 #include "shader/prog_print.h"
+#include "shader/programopt.h"
 
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
@@ -95,6 +96,9 @@ st_translate_vertex_program(struct st_context *st,
 
    memset(&vs, 0, sizeof(vs));
 
+   if (stvp->Base.IsPositionInvariant)
+      _mesa_insert_mvp_code(st->ctx, &stvp->Base);
+
    /*
     * Determine number of inputs, the mappings between VERT_ATTRIB_x
     * and TGSI generic input indexes, plus input attrib semantic info.