Merge remote branch 'origin/gallium-0.2' into gallium-0.2

39 files changed, 2498 insertions, 1096 deletions

diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 78249054f2..b439bc4059 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -274,6 +274,14 @@ draw_enable_point_sprites(struct draw_context *draw, boolean enable)
 }
 
 
+void
+draw_set_force_passthrough( struct draw_context *draw, boolean enable )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->force_passthrough = enable;
+}
+
+
 /**
  * Ask the draw module for the location/slot of the given vertex attribute in
  * a post-transformed vertex.
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index 0ab3681b64..3eeb453531 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -160,6 +160,9 @@ void draw_set_render( struct draw_context *draw,
 void draw_set_driver_clipping( struct draw_context *draw,
                                boolean bypass_clipping );
 
+void draw_set_force_passthrough( struct draw_context *draw, 
+                                 boolean enable );
+
 /*******************************************************************************
  * Draw pipeline 
  */
diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index c0cf4269db..9825e116c3 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -231,9 +231,9 @@ vbuf_set_prim( struct vbuf_stage *vbuf, uint prim )
       unsigned emit_sz = 0;
       unsigned src_buffer = 0;
       unsigned output_format;
-      unsigned src_offset = (vbuf->vinfo->src_index[i] * 4 * sizeof(float) );
+      unsigned src_offset = (vbuf->vinfo->attrib[i].src_index * 4 * sizeof(float) );
 
-      switch (vbuf->vinfo->emit[i]) {
+      switch (vbuf->vinfo->attrib[i].emit) {
       case EMIT_4F:
 	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 	 emit_sz = 4 * sizeof(float);
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 626a2e3e30..5d531146c5 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -163,12 +163,15 @@ struct draw_context
 
    struct {
       boolean bypass_clipping;
+      boolean bypass_vs;
    } driver;
 
    boolean flushing;         /**< debugging/sanity */
    boolean suspend_flushing; /**< internally set */
    boolean bypass_clipping;  /**< set if either api or driver bypass_clipping true */
 
+   boolean force_passthrough; /**< never clip or shade */
+
    /* pipe state that we need: */
    const struct pipe_rasterizer_state *rasterizer;
    struct pipe_viewport_state viewport;
@@ -193,7 +196,7 @@ struct draw_context
 
       const float (*aligned_constants)[4];
 
-      float (*aligned_constant_storage)[4];
+      const float (*aligned_constant_storage)[4];
       unsigned const_storage_size;
 
 
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 669c11c993..87ec6ae20c 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -69,26 +69,26 @@ draw_pt_arrays(struct draw_context *draw,
          return TRUE;
    }
 
-
-   if (!draw->render) {
-      opt |= PT_PIPELINE;
-   }
-
-   if (draw_need_pipeline(draw,
-                          draw->rasterizer,
-                          prim)) {
-      opt |= PT_PIPELINE;
-   }
-
-   if (!draw->bypass_clipping && !draw->pt.test_fse) {
-      opt |= PT_CLIPTEST;
+   if (!draw->force_passthrough) {
+      if (!draw->render) {
+         opt |= PT_PIPELINE;
+      }
+      
+      if (draw_need_pipeline(draw,
+                             draw->rasterizer,
+                             prim)) {
+         opt |= PT_PIPELINE;
+      }
+
+      if (!draw->bypass_clipping && !draw->pt.test_fse) {
+         opt |= PT_CLIPTEST;
+      }
+      
+      if (!draw->rasterizer->bypass_vs) {
+         opt |= PT_SHADE;
+      }
    }
-
-   if (!draw->rasterizer->bypass_vs) {
-      opt |= PT_SHADE;
-   }
-
-
+      
    if (opt == 0) 
       middle = draw->pt.middle.fetch_emit;
    else if (opt == PT_SHADE && !draw->pt.no_fse)
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
index d4eca80588..d520b05869 100644
--- a/src/gallium/auxiliary/draw/draw_pt_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -84,11 +84,11 @@ void draw_pt_emit_prepare( struct pt_emit *emit,
       unsigned emit_sz = 0;
       unsigned src_buffer = 0;
       unsigned output_format;
-      unsigned src_offset = (vinfo->src_index[i] * 4 * sizeof(float) );
+      unsigned src_offset = (vinfo->attrib[i].src_index * 4 * sizeof(float) );
 
 
          
-      switch (vinfo->emit[i]) {
+      switch (vinfo->attrib[i].emit) {
       case EMIT_4F:
 	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 	 emit_sz = 4 * sizeof(float);
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index 5a4db6cfe5..3966ad48ba 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -121,7 +121,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
    memset(&key, 0, sizeof(key));
 
    for (i = 0; i < vinfo->num_attribs; i++) {
-      const struct pipe_vertex_element *src = &draw->pt.vertex_element[vinfo->src_index[i]];
+      const struct pipe_vertex_element *src = &draw->pt.vertex_element[vinfo->attrib[i].src_index];
 
       unsigned emit_sz = 0;
       unsigned input_format = src->src_format;
@@ -129,7 +129,7 @@ static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
       unsigned input_offset = src->src_offset;
       unsigned output_format;
 
-      switch (vinfo->emit[i]) {
+      switch (vinfo->attrib[i].emit) {
       case EMIT_4F:
 	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 	 emit_sz = 4 * sizeof(float);
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 73fc70c1bc..f7e6a1a8ee 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -79,6 +79,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
    unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
    const struct vertex_info *vinfo;
    unsigned i;
+   unsigned nr_vbs = 0;
    
 
    if (!draw->render->set_primitive( draw->render, 
@@ -102,7 +103,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
 
    fse->key.viewport = !draw->identity_viewport;
    fse->key.clip = !draw->bypass_clipping;
-   fse->key.pad = 0;
+   fse->key.const_vbuffers = 0;
 
    memset(fse->key.element, 0, 
           fse->key.nr_elements * sizeof(fse->key.element[0]));
@@ -116,16 +117,23 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
        */
       fse->key.element[i].in.buffer = src->vertex_buffer_index;
       fse->key.element[i].in.offset = src->src_offset;
+      nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
    }
    
+   for (i = 0; i < 5 && i < nr_vbs; i++) {
+      if (draw->pt.vertex_buffer[i].pitch == 0)
+         fse->key.const_vbuffers |= (1<<i);
+   }
 
+   if (0) debug_printf("%s: lookup const_vbuffers: %x\n", __FUNCTION__, fse->key.const_vbuffers);
+   
    {
       unsigned dst_offset = 0;
 
       for (i = 0; i < vinfo->num_attribs; i++) {
          unsigned emit_sz = 0;
 
-         switch (vinfo->emit[i]) {
+         switch (vinfo->attrib[i].emit) {
          case EMIT_4F:
             emit_sz = 4 * sizeof(float);
             break;
@@ -153,8 +161,8 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
           * numbers, not to positions in the hw vertex description --
           * that's handled by the output_offset field.
           */
-         fse->key.element[i].out.format = vinfo->emit[i];
-         fse->key.element[i].out.vs_output = vinfo->src_index[i];
+         fse->key.element[i].out.format = vinfo->attrib[i].emit;
+         fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index;
          fse->key.element[i].out.offset = dst_offset;
       
          dst_offset += emit_sz;
@@ -162,13 +170,7 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
       }
    }
 
-
-   /* Would normally look up a vertex shader and peruse its list of
-    * varients somehow.  We omitted that step and put all the
-    * hardcoded "shaders" into an array.  We're just making the
-    * assumption that this happens to be a matching shader...  ie
-    * you're running isosurf, aren't you?
-    */
+   
    fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, 
                                          &fse->key );
 
@@ -177,18 +179,17 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
       return ;
    }
 
+   if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__, 
+                       fse->active->key.const_vbuffers);
+
    /* Now set buffer pointers:
     */
-   for (i = 0; i < num_vs_inputs; i++) {
-      unsigned buf = draw->pt.vertex_element[i].vertex_buffer_index;
-
-      fse->active->set_input( fse->active, 
-                              i, 
-                              
-                              ((const ubyte *) draw->pt.user.vbuffer[buf] + 
-                               draw->pt.vertex_buffer[buf].buffer_offset),
-                              
-                              draw->pt.vertex_buffer[buf].pitch );
+   for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+      fse->active->set_buffer( fse->active, 
+                               i, 
+                               ((const ubyte *) draw->pt.user.vbuffer[i] + 
+                                draw->pt.vertex_buffer[i].buffer_offset),
+                              draw->pt.vertex_buffer[i].pitch );
    }
 
    *max_vertices = (draw->render->max_vertex_buffer_bytes / 
diff --git a/src/gallium/auxiliary/draw/draw_vertex.c b/src/gallium/auxiliary/draw/draw_vertex.c
index 1446f785c5..3214213e44 100644
--- a/src/gallium/auxiliary/draw/draw_vertex.c
+++ b/src/gallium/auxiliary/draw/draw_vertex.c
@@ -49,7 +49,7 @@ draw_compute_vertex_size(struct vertex_info *vinfo)
 
    vinfo->size = 0;
    for (i = 0; i < vinfo->num_attribs; i++) {
-      switch (vinfo->emit[i]) {
+      switch (vinfo->attrib[i].emit) {
       case EMIT_OMIT:
          break;
       case EMIT_4UB:
@@ -81,8 +81,8 @@ draw_dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data)
    unsigned i, j;
 
    for (i = 0; i < vinfo->num_attribs; i++) {
-      j = vinfo->src_index[i];
-      switch (vinfo->emit[i]) {
+      j = vinfo->attrib[i].src_index;
+      switch (vinfo->attrib[i].emit) {
       case EMIT_OMIT:
          debug_printf("EMIT_OMIT:");
          break;
diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h
index 16c65c4317..a943607d7e 100644
--- a/src/gallium/auxiliary/draw/draw_vertex.h
+++ b/src/gallium/auxiliary/draw/draw_vertex.h
@@ -75,12 +75,41 @@ struct vertex_info
 {
    uint num_attribs;
    uint hwfmt[4];      /**< hardware format info for this format */
-   enum interp_mode interp_mode[PIPE_MAX_SHADER_INPUTS];
-   enum attrib_emit emit[PIPE_MAX_SHADER_INPUTS];   /**< EMIT_x */
-   uint src_index[PIPE_MAX_SHADER_INPUTS]; /**< map to post-xform attribs */
    uint size;          /**< total vertex size in dwords */
+   
+   /* Keep this small and at the end of the struct to allow quick
+    * memcmp() comparisons.
+    */
+   struct {
+      ubyte interp_mode:4;      /**< INTERP_x */
+      ubyte emit:4;             /**< EMIT_x */
+      ubyte src_index;          /**< map to post-xform attribs */
+   } attrib[PIPE_MAX_SHADER_INPUTS];
 };
 
+static INLINE int
+draw_vinfo_size( const struct vertex_info *a )
+{
+   return ((const char *)&a->attrib[a->num_attribs] -
+           (const char *)a);
+}
+
+static INLINE int
+draw_vinfo_compare( const struct vertex_info *a,
+                    const struct vertex_info *b )
+{
+   unsigned sizea = draw_vinfo_size( a );
+   return memcmp( a, b, sizea );
+}
+
+static INLINE void
+draw_vinfo_copy( struct vertex_info *dst,
+                 const struct vertex_info *src )
+{
+   unsigned size = draw_vinfo_size( src );
+   memcpy( dst, src, size );
+}
+
 
 
 /**
@@ -91,14 +120,15 @@ struct vertex_info
  */
 static INLINE uint
 draw_emit_vertex_attr(struct vertex_info *vinfo,
-                      enum attrib_emit emit, enum interp_mode interp,
+                      enum attrib_emit emit, 
+                      enum interp_mode interp, /* only used by softpipe??? */
                       uint src_index)
 {
    const uint n = vinfo->num_attribs;
    assert(n < PIPE_MAX_SHADER_INPUTS);
-   vinfo->emit[n] = emit;
-   vinfo->interp_mode[n] = interp;
-   vinfo->src_index[n] = src_index;
+   vinfo->attrib[n].emit = emit;
+   vinfo->attrib[n].interp_mode = interp;
+   vinfo->attrib[n].src_index = src_index;
    vinfo->num_attribs++;
    return n;
 }
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 45992d1986..68c24abad3 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -64,7 +64,7 @@ struct draw_vs_varient_key {
    unsigned nr_outputs:8;
    unsigned viewport:1;
    unsigned clip:1;
-   unsigned pad:5;
+   unsigned const_vbuffers:5;
    struct draw_varient_element element[PIPE_MAX_ATTRIBS];
 };
 
@@ -76,7 +76,7 @@ struct draw_vs_varient {
 
    struct draw_vertex_shader *vs;
 
-   void (*set_input)( struct draw_vs_varient *,
+   void (*set_buffer)( struct draw_vs_varient *,
                       unsigned i,
                       const void *ptr,
                       unsigned stride );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index a556477a76..87232865e2 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -92,9 +92,9 @@ struct x86_reg aos_get_x86( struct aos_compilation *cp,
          assert(which_reg == 1);
          offset = Offset(struct aos_machine, constants);
          break;
-      case X86_ATTRIBS:
+      case X86_BUFFERS:
          assert(which_reg == 0);
-         offset = Offset(struct aos_machine, attrib);
+         offset = Offset(struct aos_machine, buffer);
          break;
       default:
          assert(0);
@@ -196,6 +196,18 @@ static void spill( struct aos_compilation *cp, unsigned idx )
 }
 
 
+void aos_spill_all( struct aos_compilation *cp )
+{
+   unsigned i;
+
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+}
+
+
 static struct x86_reg get_xmm_writable( struct aos_compilation *cp,
                                         struct x86_reg reg )
 {
@@ -1939,6 +1951,11 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    save_fpu_state( &cp );
    set_fpu_round_nearest( &cp );
 
+   aos_init_inputs( &cp, linear );
+
+   cp.x86_reg[0] = 0;
+   cp.x86_reg[1] = 0;
+   
    /* Note address for loop jump 
     */
    label = x86_get_label(cp.func);
@@ -2018,13 +2035,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
       /* Incr index
        */   
-      if (linear) {
-         x86_inc(cp.func, cp.idx_EBX);
-      } 
-      else {
-         x86_lea(cp.func, cp.idx_EBX, x86_make_disp(cp.idx_EBX, 4));
-      }
-
+      aos_incr_inputs( &cp, linear );
    }
    /* decr count, loop if not zero
     */
@@ -2065,15 +2076,13 @@ static void vaos_set_buffer( struct draw_vs_varient *varient,
                              unsigned stride )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
-   unsigned i;
 
-   for (i = 0; i < vaos->base.key.nr_inputs; i++) {
-      if (vaos->base.key.element[i].in.buffer == buf) {
-         vaos->attrib[i].input_ptr = ((char *)ptr +
-                                      vaos->base.key.element[i].in.offset);
-         vaos->attrib[i].input_stride = stride;
-      }
+   if (buf < vaos->nr_vb) {
+      vaos->buffer[buf].base_ptr = (char *)ptr;
+      vaos->buffer[buf].stride = stride;
    }
+
+   if (0) debug_printf("%s %d/%d: %p %d\n", __FUNCTION__, buf, vaos->nr_vb, ptr, stride);
 }
 
 
@@ -2086,10 +2095,12 @@ static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
 
+   if (0) debug_printf("%s %d\n", __FUNCTION__, count);
+
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    machine->constants = vaos->draw->vs.aligned_constants;
    machine->immediates = vaos->base.vs->immediates;
-   machine->attrib = vaos->attrib;
+   machine->buffer = vaos->buffer;
 
    vaos->gen_run_elts( machine,
                        elts,
@@ -2105,10 +2116,13 @@ static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
    struct aos_machine *machine = vaos->draw->vs.aos_machine;
 
+   if (0) debug_printf("%s %d %d const: %x\n", __FUNCTION__, start, count, 
+                       vaos->base.key.const_vbuffers);
+
    machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
    machine->constants = vaos->draw->vs.aligned_constants;
    machine->immediates = vaos->base.vs->immediates;
-   machine->attrib = vaos->attrib;
+   machine->buffer = vaos->buffer;
 
    vaos->gen_run_linear( machine,
                          start,
@@ -2127,7 +2141,7 @@ static void vaos_destroy( struct draw_vs_varient *varient )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
-   FREE( vaos->attrib );
+   FREE( vaos->buffer );
 
    x86_release_func( &vaos->func[0] );
    x86_release_func( &vaos->func[1] );
@@ -2140,6 +2154,7 @@ static void vaos_destroy( struct draw_vs_varient *varient )
 static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
                                                  const struct draw_vs_varient_key *key )
 {
+   unsigned i;
    struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse);
 
    if (!vaos)
@@ -2147,17 +2162,22 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    
    vaos->base.key = *key;
    vaos->base.vs = vs;
-   vaos->base.set_input = vaos_set_buffer;
+   vaos->base.set_buffer = vaos_set_buffer;
    vaos->base.destroy = vaos_destroy;
    vaos->base.run_linear = vaos_run_linear;
    vaos->base.run_elts = vaos_run_elts;
 
    vaos->draw = vs->draw;
 
-   vaos->attrib = MALLOC( key->nr_inputs * sizeof(vaos->attrib[0]) );
-   if (!vaos->attrib)
+   for (i = 0; i < key->nr_inputs; i++) 
+      vaos->nr_vb = MAX2( vaos->nr_vb, key->element[i].in.buffer + 1 );
+
+   vaos->buffer = MALLOC( vaos->nr_vb * sizeof(vaos->buffer[0]) );
+   if (!vaos->buffer)
       goto fail;
 
+   debug_printf("nr_vb: %d const: %x\n", vaos->nr_vb, vaos->base.key.const_vbuffers);
+
 #if 0
    tgsi_dump(vs->state.tokens, 0);
 #endif
@@ -2179,8 +2199,8 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    return &vaos->base;
 
  fail:
-   if (vaos && vaos->attrib)
-      FREE(vaos->attrib);
+   if (vaos && vaos->buffer)
+      FREE(vaos->buffer);
 
    if (vaos)
       x86_release_func( &vaos->func[0] );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index 7fe6f79db0..264387517b 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -87,9 +87,10 @@ struct lit_info {
 #define MAX_SHINE_TAB    4
 #define MAX_LIT_INFO     16
 
-struct aos_attrib {
-   const void *input_ptr;
-   unsigned input_stride;
+struct aos_buffer {
+   const void *base_ptr;
+   unsigned stride;
+   void *ptr;                   /* updated per vertex */
 };
 
 
@@ -123,7 +124,7 @@ struct aos_machine {
    const float (*immediates)[4];     /* points to shader data */
    const float (*constants)[4];      /* points to draw data */
 
-   const struct aos_attrib *attrib; /* points to ? */
+   const struct aos_buffer *buffer; /* points to ? */
 };
 
 
@@ -175,12 +176,15 @@ void aos_adopt_xmm_reg( struct aos_compilation *cp,
                         unsigned idx,
                         unsigned dirty );
 
+void aos_spill_all( struct aos_compilation *cp );
+
 struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, 
                                    unsigned file,
                                    unsigned idx );
 
-boolean aos_fetch_inputs( struct aos_compilation *cp,
-                          boolean linear );
+boolean aos_init_inputs( struct aos_compilation *cp, boolean linear );
+boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear );
+boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear );
 
 boolean aos_emit_outputs( struct aos_compilation *cp );
 
@@ -210,7 +214,7 @@ do {                                                                    \
 #define X86_NULL       0
 #define X86_IMMEDIATES 1
 #define X86_CONSTANTS  2
-#define X86_ATTRIBS    3
+#define X86_BUFFERS    3
 
 struct x86_reg aos_get_x86( struct aos_compilation *cp,
                             unsigned which_reg,
@@ -232,7 +236,8 @@ struct draw_vs_varient_aos_sse {
    struct draw_vs_varient base;
    struct draw_context *draw;
 
-   struct aos_attrib *attrib;
+   struct aos_buffer *buffer;
+   unsigned nr_vb;
 
    vaos_run_linear_func gen_run_linear;
    vaos_run_elts_func gen_run_elts;
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index 26297c74f8..dd79bc799a 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -54,6 +54,7 @@ static void emit_load_R32G32B32( struct aos_compilation *cp,
 				 struct x86_reg data,
 				 struct x86_reg src_ptr )
 {
+#if 1
    sse_movss(cp->func, data, x86_make_disp(src_ptr, 8));
    /* data = z ? ? ? */
    sse_shufps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) );
@@ -62,6 +63,16 @@ static void emit_load_R32G32B32( struct aos_compilation *cp,
    /* data = ? 0 z 1 */
    sse_movlps(cp->func, data, src_ptr);
    /* data = x y z 1 */
+#else
+   sse_movups(cp->func, data, src_ptr);
+   /* data = x y z ? */
+   sse2_pshufd(cp->func, data, data, SHUF(W,X,Y,Z) );
+   /* data = ? x y z */
+   sse_movss(cp->func, data, aos_get_internal_xmm( cp, IMM_ONES ) );
+   /* data = 1 x y z */
+   sse2_pshufd(cp->func, data, data, SHUF(Y,Z,W,X) );
+   /* data = x y z 1 */
+#endif
 }
 
 static void emit_load_R32G32( struct aos_compilation *cp, 
@@ -95,28 +106,6 @@ static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp,
 
 
 
-static void get_src_ptr( struct aos_compilation *cp,
-                         struct x86_reg src,
-                         struct x86_reg elt,
-                         unsigned a )
-{
-   struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, 0, X86_ATTRIBS ), 
-                                         a * sizeof(struct aos_attrib));
-
-   struct x86_reg input_ptr = x86_make_disp(attrib, 
-                                            Offset(struct aos_attrib, input_ptr));
-
-   struct x86_reg input_stride = x86_make_disp(attrib, 
-                                               Offset(struct aos_attrib, input_stride));
-
-   /* Calculate pointer to current attrib:
-    */
-   x86_mov(cp->func, src, input_stride);
-   x86_imul(cp->func, src, elt);
-   x86_add(cp->func, src, input_ptr);
-}
-
-
 /* Extended swizzles?  Maybe later.
  */  
 static void emit_swizzle( struct aos_compilation *cp,
@@ -128,22 +117,60 @@ static void emit_swizzle( struct aos_compilation *cp,
 }
 
 
+
+static boolean get_buffer_ptr( struct aos_compilation *cp,
+                               boolean linear,
+                               unsigned buf_idx,
+                               struct x86_reg elt,
+                               struct x86_reg ptr)
+{
+   struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                      buf_idx * sizeof(struct aos_buffer));
+
+   struct x86_reg buf_stride = x86_make_disp(buf, 
+                                             Offset(struct aos_buffer, stride));
+   if (linear) {
+      struct x86_reg buf_ptr = x86_make_disp(buf, 
+                                             Offset(struct aos_buffer, ptr));
+
+
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(cp->func, ptr, buf_ptr);
+      x86_mov(cp->func, elt, buf_stride);
+      x86_add(cp->func, elt, ptr);
+      if (buf_idx == 0) sse_prefetchnta(cp->func, x86_make_disp(elt, 192));
+      x86_mov(cp->func, buf_ptr, elt);
+   }
+   else {
+      struct x86_reg buf_base_ptr = x86_make_disp(buf, 
+                                                  Offset(struct aos_buffer, base_ptr));
+
+
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(cp->func, ptr, buf_stride);
+      x86_imul(cp->func, ptr, elt);
+      x86_add(cp->func, ptr, buf_base_ptr);
+   }
+
+   cp->insn_counter++;
+
+   return TRUE;
+}
+
+
 static boolean load_input( struct aos_compilation *cp,
                            unsigned idx,
-                           boolean linear )
+                           struct x86_reg bufptr )
 {
    unsigned format = cp->vaos->base.key.element[idx].in.format;
-   struct x86_reg src = cp->tmp_EAX;
+   unsigned offset = cp->vaos->base.key.element[idx].in.offset;
    struct x86_reg dataXMM = aos_get_xmm_reg(cp);
 
    /* Figure out source pointer address:
     */
-   get_src_ptr(cp, 
-               src, 
-               linear ? cp->idx_EBX : x86_deref(cp->idx_EBX),
-               idx);
-
-   src = x86_deref(src);
+   struct x86_reg src = x86_make_disp(bufptr, offset);
 
    aos_adopt_xmm_reg( cp,
                       dataXMM,
@@ -179,20 +206,128 @@ static boolean load_input( struct aos_compilation *cp,
    return TRUE;
 }
 
-
-boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
+static boolean load_inputs( struct aos_compilation *cp,
+                            unsigned buffer,
+                            struct x86_reg ptr )
 {
    unsigned i;
-   
+
    for (i = 0; i < cp->vaos->base.key.nr_inputs; i++) {
-      if (!load_input( cp, i, linear ))
-         return FALSE;
-      cp->insn_counter++;
+      if (cp->vaos->base.key.element[i].in.buffer == buffer) {
+
+         if (!load_input( cp, i, ptr ))
+            return FALSE;
+
+         cp->insn_counter++;
+      }
+   }
+   
+   return TRUE;
+}
+
+boolean aos_init_inputs( struct aos_compilation *cp, boolean linear )
+{
+   unsigned i;
+   for (i = 0; i < cp->vaos->nr_vb; i++) {
+      struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                         i * sizeof(struct aos_buffer));
+
+      struct x86_reg buf_base_ptr = x86_make_disp(buf, 
+                                                  Offset(struct aos_buffer, base_ptr));
+
+      if (cp->vaos->base.key.const_vbuffers & (1<<i)) {
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         x86_mov(cp->func, ptr, buf_base_ptr);
+
+         /* Load all inputs for this constant vertex buffer
+          */
+         load_inputs( cp, i, x86_deref(ptr) );
+         
+         /* Then just force them out to aos_machine.input[]
+          */
+         aos_spill_all( cp );
+
+      }
+      else if (linear) {
+
+         struct x86_reg elt = cp->idx_EBX;
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         struct x86_reg buf_stride = x86_make_disp(buf, 
+                                                   Offset(struct aos_buffer, stride));
+
+         struct x86_reg buf_ptr = x86_make_disp(buf, 
+                                                Offset(struct aos_buffer, ptr));
+
+
+         /* Calculate pointer to current attrib:
+          */
+         x86_mov(cp->func, ptr, buf_stride);
+         x86_imul(cp->func, ptr, elt);
+         x86_add(cp->func, ptr, buf_base_ptr);
+
+
+         /* In the linear case, keep the buffer pointer instead of the
+          * index number.
+          */
+         if (cp->vaos->nr_vb == 1) 
+            x86_mov( cp->func, elt, ptr );
+         else
+            x86_mov( cp->func, buf_ptr, ptr );
+
+         cp->insn_counter++;
+      }
+   }
+
+   return TRUE;
+}
+
+boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
+{
+   unsigned j;
+
+   for (j = 0; j < cp->vaos->nr_vb; j++) {
+      if (cp->vaos->base.key.const_vbuffers & (1<<j)) {
+         /* just retreive pre-transformed input */
+      }
+      else if (linear && cp->vaos->nr_vb == 1) {
+         load_inputs( cp, 0, cp->idx_EBX );
+      }
+      else {
+         struct x86_reg elt = linear ? cp->idx_EBX : x86_deref(cp->idx_EBX);
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         if (!get_buffer_ptr( cp, linear, j, elt, ptr ))
+            return FALSE;
+
+         if (!load_inputs( cp, j, ptr ))
+            return FALSE;
+      }
    }
 
    return TRUE;
 }
 
+boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear )
+{
+   if (linear && cp->vaos->nr_vb == 1) {
+      struct x86_reg stride = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                            (0 * sizeof(struct aos_buffer) + 
+                                             Offset(struct aos_buffer, stride)));
+
+      x86_add(cp->func, cp->idx_EBX, stride);
+      sse_prefetchnta(cp->func, x86_make_disp(cp->idx_EBX, 192));
+   }
+   else if (linear) {
+      /* Nothing to do */
+   } 
+   else {
+      x86_lea(cp->func, cp->idx_EBX, x86_make_disp(cp->idx_EBX, 4));
+   }
+
+   return TRUE;
+}
 
 
 
@@ -203,7 +338,7 @@ static void emit_store_R32G32B32A32( struct aos_compilation *cp,
 				     struct x86_reg dst_ptr,
 				     struct x86_reg dataXMM )
 {
-   sse_movups(cp->func, dst_ptr, dataXMM);
+   sse_movaps(cp->func, dst_ptr, dataXMM);
 }
 
 static void emit_store_R32G32B32( struct aos_compilation *cp, 
@@ -306,7 +441,7 @@ boolean aos_emit_outputs( struct aos_compilation *cp )
 
       if (data.file != file_XMM) {
          struct x86_reg tmp = aos_get_xmm_reg( cp );
-         sse_movups(cp->func, tmp, data);
+         sse_movaps(cp->func, tmp, data);
          data = tmp;
       }
       
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
index 2ce30b9a02..727977bc3a 100644
--- a/src/gallium/auxiliary/draw/draw_vs_llvm.c
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -32,6 +32,7 @@
   *   Brian Paul
   */
 
+#include "util/u_memory.h"
 #include "pipe/p_shader_tokens.h"
 #include "draw_private.h"
 #include "draw_context.h"
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 0efabd9de8..b11ae31662 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -37,7 +37,7 @@
 
 #include "draw_vs.h"
 
-#if defined(PIPE_ARCH_X86)
+#if defined(PIPE_ARCH_X86) && defined(PIPE_ARCH_SSE)
 
 #include "pipe/p_shader_tokens.h"
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 4daf05dae7..7ee567d478 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -64,10 +64,10 @@ struct draw_vs_varient_generic {
 
 
 
-static void vsvg_set_input( struct draw_vs_varient *varient,
-                            unsigned buffer,
-                            const void *ptr,
-                            unsigned stride )
+static void vsvg_set_buffer( struct draw_vs_varient *varient,
+                             unsigned buffer,
+                             const void *ptr,
+                             unsigned stride )
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 
@@ -265,7 +265,7 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
 
    vsvg->base.key = *key;
    vsvg->base.vs = vs;
-   vsvg->base.set_input     = vsvg_set_input;
+   vsvg->base.set_buffer    = vsvg_set_buffer;
    vsvg->base.run_elts      = vsvg_run_elts;
    vsvg->base.run_linear    = vsvg_run_linear;
    vsvg->base.destroy       = vsvg_destroy;
diff --git a/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
index 0fc5c4ec5c..fcc5c05794 100644
--- a/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
@@ -1,140 +1,140 @@
 static const unsigned char llvm_builtins_data[] = {
-0x42,0x43,0xc0,0xde,0x21,0x0c,0x00,0x00,0x29,0x02,0x00,0x00,0x01,0x10,0x00,0x00,
+0x42,0x43,0xc0,0xde,0x21,0x0c,0x00,0x00,0x27,0x02,0x00,0x00,0x01,0x10,0x00,0x00,
 0x10,0x00,0x00,0x00,0x07,0x81,0x23,0x91,0x41,0xc8,0x04,0x49,0x06,0x10,0x32,0x39,
 0x92,0x01,0x84,0x0c,0x25,0x05,0x08,0x19,0x1e,0x04,0x8b,0x62,0x80,0x14,0x45,0x02,
 0x42,0x92,0x0b,0x42,0xa4,0x10,0x32,0x14,0x38,0x08,0x18,0x49,0x0a,0x32,0x44,0x24,
 0x48,0x0a,0x90,0x21,0x23,0x44,0x72,0x80,0x8c,0x14,0x21,0x86,0x0a,0x8a,0x0a,0x64,
-0x0c,0x1f,0x00,0x00,0x49,0x18,0x00,0x00,0x02,0x00,0x00,0x00,0x0b,0x04,0x00,0x0c,
-0x00,0x00,0x00,0x00,0x51,0x20,0x00,0x00,0x12,0x00,0x00,0x00,0x32,0x22,0x48,0x09,
-0x20,0x65,0x82,0x84,0x00,0x26,0x45,0x48,0x05,0x09,0x26,0x45,0xc6,0x05,0x42,0x52,
-0x26,0x08,0xae,0x19,0x80,0x61,0x04,0x02,0x98,0x23,0x00,0x83,0x29,0x80,0x21,0x00,
-0xb2,0x73,0x04,0x01,0x51,0x8a,0xf4,0x08,0x92,0xa4,0x39,0x47,0x80,0x50,0x2b,0x03,
-0x00,0xa0,0x08,0x21,0x5c,0x46,0x2b,0x44,0x08,0x21,0xd4,0x40,0x14,0x01,0x80,0x11,
-0x80,0x22,0x88,0x00,0x13,0xa2,0x74,0xb0,0x03,0x3c,0xb0,0x83,0x36,0x80,0x87,0x71,
-0x68,0x03,0x76,0x48,0x07,0x77,0xa8,0x07,0x7c,0x68,0x83,0x73,0x70,0x87,0x7a,0xd8,
-0x70,0x0f,0xe5,0xd0,0x06,0xf0,0xa0,0x07,0x73,0x20,0x07,0x7a,0x30,0x07,0x72,0xa0,
-0x07,0x73,0x20,0x07,0x6d,0x90,0x0e,0x71,0xa0,0x07,0x78,0xa0,0x07,0x78,0xd0,0x06,
-0xe9,0x80,0x07,0x7a,0x80,0x07,0x7a,0x80,0x07,0x6d,0x90,0x0e,0x71,0x60,0x07,0x7a,
-0x10,0x07,0x76,0xa0,0x07,0x71,0x60,0x07,0x6d,0x90,0x0e,0x73,0x20,0x07,0x7a,0x30,
-0x07,0x72,0xa0,0x07,0x73,0x20,0x07,0x6d,0x90,0x0e,0x76,0x40,0x07,0x7a,0x30,0x07,
-0x72,0xa0,0x07,0x76,0x40,0x07,0x6d,0x60,0x0e,0x73,0x20,0x07,0x7a,0x30,0x07,0x72,
-0xa0,0x07,0x73,0x20,0x07,0x6d,0x60,0x0e,0x76,0x40,0x07,0x7a,0x30,0x07,0x72,0xa0,
-0x07,0x76,0x40,0x07,0x6d,0x60,0x0f,0x76,0x40,0x07,0x7a,0x60,0x07,0x74,0xa0,0x07,
-0x76,0x40,0x07,0x6d,0x60,0x0f,0x71,0x20,0x07,0x78,0xa0,0x07,0x71,0x20,0x07,0x78,
-0xa0,0x07,0x71,0x20,0x07,0x78,0xd0,0x06,0xe1,0x00,0x07,0x7a,0x00,0x07,0x7a,0x60,
-0x07,0x74,0xd0,0x06,0xe6,0x80,0x07,0x70,0xa0,0x07,0x71,0x20,0x07,0x78,0xa0,0x07,
-0x71,0x20,0x07,0x78,0xa0,0xf3,0x40,0x88,0x04,0x32,0x32,0x02,0x04,0x20,0x76,0x46,
-0xfc,0x6c,0x48,0x92,0x00,0x40,0x00,0x00,0x00,0x00,0x0c,0x49,0x12,0x20,0x00,0x00,
-0x00,0x00,0x80,0x21,0x89,0x02,0x00,0x01,0x00,0x00,0x00,0x30,0x24,0x59,0x00,0x20,
-0x08,0x00,0x00,0x00,0x86,0x24,0x0a,0x00,0x04,0x00,0x00,0x00,0xc0,0x90,0x84,0x01,
-0x02,0x00,0x00,0x00,0x00,0x18,0x92,0x1c,0x40,0x00,0x00,0x00,0x00,0x00,0x43,0x12,
-0x05,0x00,0x02,0x00,0x00,0x00,0x60,0x48,0x72,0x00,0x01,0x00,0x00,0x00,0x00,0x0c,
-0x49,0x14,0x00,0x08,0x00,0x00,0x00,0x80,0x21,0x49,0x01,0x00,0x41,0x00,0x00,0x00,
-0x90,0x05,0x02,0x00,0x10,0x00,0x00,0x00,0x32,0x1e,0x98,0x10,0x19,0x11,0x4c,0x90,
+0x0c,0x1f,0x00,0x00,0x49,0x18,0x00,0x00,0x03,0x00,0x00,0x00,0x0b,0x84,0xff,0xff,
+0xff,0xff,0x1f,0xc0,0x00,0x00,0x00,0x00,0x51,0x20,0x00,0x00,0x12,0x00,0x00,0x00,
+0x32,0x22,0x48,0x09,0x20,0x65,0x82,0x84,0x00,0x26,0x45,0x48,0x05,0x09,0x26,0x45,
+0xc6,0x05,0x42,0x52,0x26,0x08,0xae,0x19,0x80,0x61,0x04,0x02,0x98,0x23,0x00,0x83,
+0x29,0x80,0x21,0x00,0xb2,0x73,0x04,0x01,0x51,0x8a,0xf4,0x08,0x92,0xa4,0x39,0x47,
+0x80,0x50,0x2b,0x03,0x00,0xa0,0x08,0x21,0x5c,0x46,0x2b,0x44,0x08,0x21,0xd4,0x40,
+0x14,0x01,0x80,0x11,0x80,0x22,0x88,0x00,0x13,0x30,0x7c,0xc0,0x03,0x3b,0xf8,0x05,
+0x3b,0xa0,0x83,0x36,0xa8,0x07,0x77,0x58,0x07,0x77,0x78,0x87,0x7b,0x70,0x87,0x36,
+0x60,0x87,0x74,0x70,0x87,0x7a,0xc0,0x87,0x36,0x38,0x07,0x77,0xa8,0x87,0x0d,0xf7,
+0x50,0x0e,0x6d,0x00,0x0f,0x7a,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60,
+0x07,0x74,0xd0,0x06,0xe9,0x10,0x07,0x7a,0x80,0x07,0x7a,0x80,0x07,0x6d,0x90,0x0e,
+0x78,0xa0,0x07,0x78,0xa0,0x07,0x78,0xd0,0x06,0xe9,0x10,0x07,0x76,0xa0,0x07,0x71,
+0x60,0x07,0x7a,0x10,0x07,0x76,0xd0,0x06,0xe9,0x30,0x07,0x72,0xa0,0x07,0x73,0x20,
+0x07,0x7a,0x30,0x07,0x72,0xd0,0x06,0xe9,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,
+0x7a,0x60,0x07,0x74,0xd0,0x06,0xe6,0x30,0x07,0x72,0xa0,0x07,0x73,0x20,0x07,0x7a,
+0x30,0x07,0x72,0xd0,0x06,0xe6,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60,
+0x07,0x74,0xd0,0x06,0xf6,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60,0x07,
+0x74,0xd0,0x06,0xf6,0x10,0x07,0x72,0x80,0x07,0x7a,0x10,0x07,0x72,0x80,0x07,0x7a,
+0x10,0x07,0x72,0x80,0x07,0x6d,0x10,0x0e,0x70,0xa0,0x07,0x70,0xa0,0x07,0x76,0x40,
+0x07,0x6d,0x60,0x0e,0x78,0x00,0x07,0x7a,0x10,0x07,0x72,0x80,0x07,0x7a,0x10,0x07,
+0x72,0x80,0x07,0x3a,0x0f,0x84,0x48,0x20,0x23,0x24,0x40,0x00,0x62,0x67,0x88,0x9f,
+0x19,0x92,0x24,0x00,0x10,0x04,0x00,0x00,0x00,0x43,0x92,0x04,0x08,0x00,0x00,0x00,
+0x00,0x60,0x48,0xa2,0x00,0x40,0x10,0x00,0x00,0x00,0x0c,0x49,0x16,0x00,0x08,0x02,
+0x00,0x00,0x80,0x21,0x89,0x02,0x00,0x41,0x00,0x00,0x00,0x30,0x24,0x61,0x80,0x00,
+0x00,0x00,0x00,0x00,0x86,0x24,0x07,0x10,0x00,0x00,0x00,0x00,0xc0,0x90,0x44,0x01,
+0x80,0x20,0x00,0x00,0x00,0x18,0x92,0x1c,0x40,0x00,0x00,0x00,0x00,0x00,0x43,0x12,
+0x05,0x00,0x82,0x00,0x00,0x00,0x60,0x48,0x52,0x00,0x40,0x10,0x00,0x00,0x00,0x64,
+0x81,0x00,0x00,0x00,0x10,0x00,0x00,0x00,0x32,0x1e,0x98,0x10,0x19,0x11,0x4c,0x90,
 0x8c,0x09,0x26,0x47,0xc6,0x04,0x43,0x8a,0x8a,0x59,0x8b,0x43,0x50,0xd2,0x09,0x02,
 0x81,0xd2,0x73,0x50,0xc9,0x0c,0x2a,0x99,0x41,0x25,0x33,0xa8,0x64,0x56,0x28,0x66,
 0x2d,0x0e,0x41,0xcf,0x2a,0x15,0x04,0x4a,0xcf,0x41,0x25,0x33,0xa8,0x64,0x06,0x95,
 0xcc,0xa0,0x92,0x59,0x01,0x00,0x00,0x00,0x53,0x82,0x26,0x0c,0x04,0x00,0x00,0x00,
 0x22,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x05,0x00,0x00,0x00,
 0x04,0xc6,0x08,0x40,0x10,0x04,0xe1,0x70,0x18,0x23,0x00,0x41,0x10,0x84,0xc3,0x60,
-0x04,0x00,0x00,0x00,0x93,0x0c,0xce,0x43,0x4c,0x31,0x3c,0x8e,0x34,0xc9,0x30,0x41,
-0xc2,0x14,0x03,0x34,0x51,0x93,0x0c,0x4d,0x44,0x4c,0x31,0x44,0x8d,0x35,0x56,0x01,
-0x04,0xc3,0x55,0x21,0x16,0x0e,0x04,0x00,0x0f,0x00,0x00,0x00,0x46,0x41,0x08,0xcc,
-0x73,0x9b,0x05,0x21,0x30,0xcf,0x6e,0x18,0x84,0x00,0x2c,0x8b,0x35,0x04,0x80,0x39,
-0x04,0x81,0x5d,0x20,0x80,0x0f,0x0c,0x43,0xe4,0xd3,0x36,0x81,0x04,0x3e,0x30,0x0c,
-0x91,0x4f,0x5b,0x05,0x12,0xf8,0xc0,0x30,0x44,0x7e,0x7d,0x00,0x05,0xd1,0x4c,0x11,
-0x66,0x12,0x83,0xc0,0x3c,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
-0x2a,0x00,0x00,0x00,0x13,0x04,0x43,0x2c,0x10,0x00,0x00,0x00,0x08,0x00,0x00,0x00,
+0x04,0x00,0x00,0x00,0xc3,0x0d,0xce,0x43,0x4c,0x37,0x3c,0x8e,0x34,0xdc,0x30,0x41,
+0xc2,0x74,0x03,0x34,0x51,0xc3,0x0d,0x4d,0x44,0x4c,0x37,0x44,0x8d,0x35,0x56,0x01,
+0x04,0xc3,0x55,0x21,0x16,0x0e,0x04,0x00,0x0f,0x00,0x00,0x00,0xd6,0x10,0x00,0xe6,
+0x10,0x04,0x76,0x81,0x00,0x3e,0x30,0x0c,0x91,0x4f,0x1b,0x05,0x21,0x30,0x8f,0x6d,
+0x13,0x48,0xe0,0x03,0xc3,0x10,0xf9,0xb4,0x55,0x20,0x81,0x0f,0x0c,0x43,0xe4,0xd7,
+0x66,0x41,0x08,0xcc,0xa3,0x1f,0x40,0x41,0x34,0x53,0x84,0x99,0xc4,0x20,0x30,0x8f,
+0x61,0x10,0x02,0xb0,0x2c,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x27,0x00,0x00,0x00,0x13,0x04,0x43,0x2c,0x10,0x00,0x00,0x00,0x08,0x00,0x00,0x00,
 0x24,0x8a,0xa0,0x0c,0x46,0x00,0x4a,0x80,0xc2,0x1c,0x84,0x55,0x55,0xd6,0x1c,0x84,
 0x45,0x51,0x16,0x81,0x19,0x80,0x11,0x80,0x31,0x02,0x10,0x04,0x41,0xfc,0x03,0x00,
-0x63,0x08,0x0d,0x34,0xc9,0x70,0x55,0xc2,0x2c,0x43,0x20,0x60,0x73,0x0c,0xd3,0x15,
+0x63,0x08,0x0d,0x34,0xdc,0x70,0x55,0xc2,0x2c,0x43,0x20,0x60,0x73,0x0c,0xd3,0x15,
 0x8d,0x21,0x34,0xd1,0x18,0x42,0xf3,0x8c,0x55,0x00,0x81,0xa0,0x6d,0x73,0x0c,0x19,
-0xe7,0x60,0x87,0x52,0x38,0x10,0x00,0x00,0x13,0x00,0x00,0x00,0x17,0x60,0x20,0xc5,
-0x74,0x10,0x8d,0x65,0x14,0x13,0xf3,0xd4,0xb4,0x6d,0x14,0x13,0xf3,0xd4,0xb8,0x69,
-0x14,0x13,0xf3,0xd4,0xb6,0x75,0x14,0x13,0xf3,0xd4,0xba,0x35,0x0c,0x13,0xf3,0x9c,
-0x80,0xe4,0x36,0x48,0x81,0x10,0xc3,0x4a,0x4c,0x54,0xd4,0x6c,0x8b,0x23,0x28,0x76,
-0x41,0x4c,0xcc,0xa3,0x1b,0x07,0x21,0x00,0xcb,0x72,0x00,0x05,0xd1,0x4c,0x11,0x66,
-0x18,0x83,0xc0,0x3c,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
-0x81,0x00,0x00,0x00,0x13,0x04,0x4d,0x2c,0x10,0x00,0x00,0x00,0x04,0x00,0x00,0x00,
-0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x91,0x11,0x00,0x00,0x00,
-0x63,0x08,0x4d,0x64,0x16,0xc1,0x49,0x86,0xab,0x22,0x66,0x19,0x02,0x01,0x1b,0x43,
-0x70,0xa2,0x59,0x82,0x61,0x0c,0xe1,0x89,0x66,0x09,0x86,0x81,0x0a,0x20,0x0b,0x34,
-0x61,0x8e,0x81,0xda,0xa2,0x31,0x84,0x46,0xb2,0x8e,0xe0,0x24,0x83,0x57,0x11,0xb3,
-0x0c,0x44,0xf1,0x8d,0x21,0x38,0xd2,0x2c,0x81,0x31,0x86,0xf0,0x48,0xb3,0x04,0xc6,
-0x40,0x05,0x00,0x06,0x44,0x18,0x14,0x73,0x0c,0x9c,0x18,0x48,0x63,0x08,0xcd,0x64,
-0x64,0x40,0x70,0x92,0xa1,0x0c,0x2a,0x62,0x96,0xe1,0x40,0xcc,0x60,0x0c,0xc1,0x99,
-0x66,0x09,0x92,0x31,0x84,0x67,0x9a,0x25,0x48,0x06,0x2a,0x80,0x33,0x38,0xd0,0x00,
-0x99,0x63,0x18,0x83,0x34,0x98,0xc6,0x10,0x1a,0xc8,0xd6,0x80,0xe0,0x24,0x03,0x1b,
-0x54,0xc4,0x2c,0x83,0xb2,0xb4,0xc1,0x18,0x82,0x03,0xcd,0x12,0x30,0x63,0x08,0x0f,
-0x34,0x4b,0xc0,0x0c,0x54,0x00,0x6e,0xa0,0xbc,0xc1,0x32,0xc7,0xa0,0x06,0x70,0x00,
-0x61,0x1c,0x84,0x03,0x01,0x00,0x00,0x00,0x4e,0x00,0x00,0x00,0x76,0x52,0x4c,0xcc,
-0x73,0xd3,0x24,0x05,0x64,0xec,0xcd,0x8d,0xcc,0xe5,0x87,0x46,0xc6,0x50,0x8a,0x89,
-0x79,0xee,0xdb,0x54,0x8a,0x89,0x79,0xee,0xdd,0x1a,0x88,0x89,0x79,0x68,0x73,0x20,
-0x26,0xe6,0xa9,0xed,0x81,0x98,0x98,0xc7,0x36,0x0b,0x62,0x62,0x9e,0xdb,0x32,0x88,
-0x89,0x79,0x72,0xd3,0x20,0x26,0xe6,0xd9,0x8d,0x83,0x98,0x98,0xa7,0xb7,0x95,0x62,
-0x62,0x9e,0xbb,0x27,0x2d,0x20,0x63,0x6f,0x6e,0x64,0x2e,0x3a,0x34,0x35,0x56,0x62,
-0x08,0x4e,0x53,0xd9,0xba,0xb5,0x14,0x02,0xf3,0xe0,0xf5,0x25,0x2c,0x82,0xd3,0x0c,
-0xbe,0xe0,0x34,0xd3,0x8d,0x9b,0x88,0x21,0x38,0xcd,0x60,0xd7,0x24,0x01,0x63,0xec,
-0xcd,0x8d,0xcc,0x45,0x87,0x44,0x80,0x8c,0xbd,0xb9,0x91,0xb9,0xfc,0xc4,0xd0,0x90,
-0x02,0x8c,0xb1,0x37,0x37,0x32,0x97,0x1f,0x73,0x29,0x26,0xe6,0xc1,0x71,0x7b,0x29,
-0x26,0xe6,0xc1,0x77,0xfb,0x28,0x04,0xe6,0xa9,0x6f,0x52,0x01,0x32,0xf6,0xe6,0x46,
-0xe6,0xa2,0x13,0x73,0x63,0x18,0x83,0xc0,0x3c,0xb6,0x41,0x08,0x4e,0x33,0x58,0x47,
-0x31,0x31,0x4f,0x5d,0x1f,0xc3,0x22,0x38,0xcd,0xe0,0x0b,0x4e,0x33,0xe1,0xbc,0xa5,
-0x18,0x82,0xd3,0x0c,0x77,0x6e,0x20,0xc5,0xc4,0x3c,0xb5,0x4e,0x3a,0x40,0xc6,0xde,
-0xdc,0xc8,0x5c,0x7e,0x64,0x70,0x2c,0xa4,0x98,0x98,0xa7,0xee,0xed,0x82,0x10,0x9c,
-0xa6,0xba,0x81,0x44,0x70,0x9a,0xc1,0x17,0x9c,0x66,0x32,0x93,0x42,0x60,0x1e,0x7b,
-0xb7,0x98,0x62,0x62,0x9e,0xbc,0x36,0x16,0x43,0x70,0x9a,0x0a,0xa7,0x6d,0xa4,0x98,
-0x98,0xc7,0xbe,0x8d,0xa4,0x98,0x98,0xc7,0xce,0x0d,0xc6,0x10,0x9c,0x66,0xc0,0x7b,
-0x12,0x02,0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x33,0x13,0x73,0x06,0x8b,0xe0,0x34,0x83,
-0x2f,0x38,0xcd,0x64,0xd3,0xe6,0x61,0x08,0x4e,0x53,0xd5,0xf6,0x01,0x14,0x44,0x33,
-0x45,0x18,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x4a,0x00,0x00,0x00,
-0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x07,0x00,0x00,0x00,0x24,0xca,0x60,0x04,
-0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0xb9,0x61,0x0c,0x04,0x10,0x1e,0xe1,0x19,0xc6,
-0x40,0x02,0xe1,0x11,0x1e,0x00,0x00,0x00,0x63,0x08,0xcd,0x63,0x15,0xc1,0x31,0x84,
-0x06,0xb2,0x8b,0xe0,0x18,0x42,0x13,0x59,0x46,0x70,0x0c,0xa1,0x71,0x6c,0x23,0x38,
-0x16,0x02,0x04,0xc7,0x64,0x61,0x1a,0x37,0x16,0x01,0x04,0x48,0x35,0xc7,0x20,0x79,
-0xcf,0x58,0x04,0x10,0x20,0xd5,0x1c,0xc3,0x07,0x06,0xd0,0x58,0x04,0x10,0x20,0xd5,
-0x1c,0x43,0x18,0x88,0x41,0x34,0x16,0x01,0x04,0x48,0x35,0xc7,0x30,0x06,0x64,0xe0,
-0x98,0x47,0xd0,0xc0,0x80,0xa0,0x89,0x01,0x41,0x23,0x03,0x82,0x63,0x21,0x40,0x70,
-0x50,0x66,0x70,0x06,0x68,0x90,0x06,0x58,0x06,0xe1,0x40,0x00,0x25,0x00,0x00,0x00,
-0x56,0x52,0x4c,0xcc,0x73,0xd3,0x56,0x41,0x4c,0xcc,0x53,0xdb,0x05,0x31,0x31,0xcf,
-0x6d,0x19,0xc4,0xc4,0x3c,0xba,0x6d,0x10,0x13,0xf3,0xf4,0xd6,0x41,0x08,0xc0,0xb2,
-0x18,0x46,0x21,0x38,0x4d,0x85,0x9b,0x46,0x21,0x38,0x4d,0xb5,0x9b,0x8a,0x21,0x00,
-0xcb,0x82,0xdf,0x66,0x62,0x08,0x4e,0x53,0xdd,0xb7,0x9d,0x18,0x82,0xd3,0x54,0xb7,
-0x6e,0x28,0x86,0xe0,0x34,0xd5,0xdd,0xdb,0x47,0x31,0x31,0x4f,0x9d,0x9b,0x87,0x21,
-0x00,0xcb,0x52,0xdf,0x06,0x62,0x08,0xc0,0xb2,0xd4,0xbc,0x59,0x10,0x82,0xd3,0x54,
-0x96,0x62,0x08,0x4e,0x53,0xe1,0xb6,0x85,0x14,0x13,0xf3,0xd8,0xb4,0x8d,0x14,0x13,
-0xf3,0xd8,0xb9,0x89,0x18,0x02,0xb0,0x2c,0xf6,0x6d,0x24,0x86,0x00,0x2c,0x8b,0xcd,
-0x1b,0x87,0x21,0x38,0x4d,0x55,0xd3,0xd6,0x30,0x54,0xc0,0x72,0x00,0x05,0xd1,0x4c,
-0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x19,0x00,0x00,0x00,
-0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0x4a,0x60,0x04,
-0x80,0xc2,0x0c,0x00,0x00,0x00,0x00,0x00,0x63,0x08,0xcd,0x33,0x16,0x01,0x04,0x48,
-0x34,0xc7,0x00,0x49,0xcf,0x58,0x04,0x10,0x28,0xd1,0x1c,0xc3,0x44,0x39,0x58,0x85,
-0x03,0x01,0x00,0x00,0x0a,0x00,0x00,0x00,0x16,0x41,0x4c,0xcc,0x63,0xdb,0x04,0x31,
-0x31,0x4f,0x6e,0x0d,0x43,0x05,0x2c,0x07,0x50,0x10,0xcd,0x14,0x61,0x56,0x41,0x4c,
-0xcc,0xd3,0x1b,0x45,0x21,0x00,0xcb,0xb2,0x9b,0x04,0x21,0x00,0xcb,0x02,0x00,0x00,
-0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x1b,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,
-0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,
-0xc2,0x0c,0x00,0x00,0x63,0x08,0xcd,0x33,0x16,0x01,0x04,0xca,0x34,0xc7,0x20,0x51,
-0xcf,0x1c,0x43,0x45,0x41,0x73,0x0c,0x16,0x15,0xcd,0x31,0x5c,0x94,0x83,0x58,0x38,
-0x10,0x00,0x00,0x00,0x0b,0x00,0x00,0x00,0x86,0x51,0x4c,0xcc,0x53,0xe7,0x76,0x51,
-0x4c,0xcc,0x53,0xdb,0x36,0x41,0x4c,0xcc,0x63,0x5b,0x05,0x31,0x31,0x8f,0x6e,0x0d,
-0x43,0x05,0x2c,0x66,0x41,0x4c,0xcc,0xd3,0x1f,0x40,0x41,0x34,0x53,0x84,0x19,0x05,
-0x21,0x00,0xcb,0x02,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x2f,0x00,0x00,0x00,
-0x13,0x04,0x45,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0xca,0xa0,0x04,
-0x46,0x00,0x8a,0x80,0xc0,0x08,0x00,0x00,0x63,0x08,0x0d,0x34,0xc9,0x30,0x49,0xc4,
-0x2c,0x03,0x11,0x50,0x63,0x08,0xcd,0x33,0xc9,0x50,0x49,0xc4,0x2c,0x03,0x21,0x58,
-0x63,0x08,0x4d,0x34,0xc9,0x70,0x49,0xc4,0x2c,0x03,0x31,0x60,0x63,0x08,0x8d,0x33,
-0xc9,0x90,0x49,0x84,0x69,0x22,0x70,0xc3,0x27,0x1c,0x08,0x00,0x1a,0x00,0x00,0x00,
-0x96,0x51,0x4c,0xcc,0x53,0xdf,0x66,0x41,0x08,0xcc,0x83,0xdb,0x04,0x31,0x31,0x4f,
-0x6d,0x15,0xc4,0xc4,0x3c,0xb7,0x61,0x10,0x02,0xf3,0xf0,0x47,0x20,0xb9,0x0d,0x52,
-0x20,0xc4,0xb0,0x12,0x13,0x15,0x35,0xdb,0xe2,0x08,0x8a,0x5d,0x10,0x13,0xf3,0xec,
-0x37,0x90,0x2c,0x4e,0xf4,0x47,0x87,0x54,0xd7,0x17,0x70,0x2c,0x4e,0xf4,0x47,0x87,
-0x74,0x02,0xc8,0xe2,0x44,0x7f,0x74,0x48,0xb9,0x69,0x14,0x02,0xf3,0xd4,0xb8,0x6d,
-0x18,0x11,0x31,0x55,0xc0,0x62,0x0d,0x43,0x05,0x2c,0x07,0x50,0x10,0xcd,0x14,0x61,
-0x46,0x31,0x08,0xcc,0x03,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x71,0x20,0x00,0x00,
-0x12,0x00,0x00,0x00,0x66,0x40,0x54,0x82,0x23,0x59,0xc2,0x20,0x09,0x92,0x1d,0x18,
-0x4f,0x84,0x34,0x53,0x61,0x03,0xc4,0xe3,0x58,0x85,0x05,0x14,0xbe,0x34,0x45,0xb5,
-0x21,0x10,0x82,0x23,0x15,0x46,0x30,0x2c,0xc8,0x64,0x02,0x06,0xf0,0x3c,0x91,0x73,
-0x19,0x00,0xe1,0x4b,0x53,0x64,0x0a,0x84,0x84,0x34,0x85,0x31,0x10,0x0a,0xb2,0x3c,
-0x56,0x30,0x08,0xcc,0x63,0x0b,0x44,0x25,0x21,0x0d,0x00,0x00,0x00,0x00,0x00,0x00};
+0xe7,0x60,0x87,0x52,0x38,0x10,0x00,0x00,0x10,0x00,0x00,0x00,0x27,0x50,0x20,0x05,
+0xd1,0x0c,0x17,0x60,0x20,0xc5,0x74,0x10,0x8d,0x65,0x14,0x13,0xf3,0xd4,0xb4,0x6d,
+0x14,0x13,0xf3,0xd4,0xb8,0x69,0x14,0x13,0xf3,0xd4,0xb6,0x75,0x14,0x13,0xf3,0xd4,
+0xba,0x35,0x0c,0x13,0xf3,0xd8,0x05,0x31,0x31,0x8f,0x6e,0x1c,0x84,0x00,0x2c,0xcb,
+0x01,0x14,0x44,0x33,0x45,0x98,0x61,0x0c,0x02,0xf3,0x00,0x00,0x00,0x00,0x00,0x00,
+0x61,0x20,0x00,0x00,0x81,0x00,0x00,0x00,0x13,0x04,0x4d,0x2c,0x10,0x00,0x00,0x00,
+0x04,0x00,0x00,0x00,0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x91,
+0x11,0x00,0x00,0x00,0x63,0x08,0x4d,0x64,0x16,0xc1,0xe1,0x86,0xab,0x22,0x66,0x19,
+0x02,0x01,0x1b,0x43,0x70,0xa2,0x59,0x82,0x61,0x0c,0xe1,0x89,0x66,0x09,0x86,0x81,
+0x0a,0x20,0x0b,0x34,0x61,0x8e,0x81,0xda,0xa2,0x31,0x84,0x46,0xb2,0x8e,0xe0,0x70,
+0x83,0x57,0x11,0xb3,0x0c,0x44,0xf1,0x8d,0x21,0x38,0xd2,0x2c,0x81,0x31,0x86,0xf0,
+0x48,0xb3,0x04,0xc6,0x40,0x05,0x00,0x06,0x44,0x18,0x14,0x73,0x0c,0x9c,0x18,0x48,
+0x63,0x08,0xcd,0x64,0x64,0x40,0x70,0xb8,0xa1,0x0c,0x2a,0x62,0x96,0xe1,0x40,0xcc,
+0x60,0x0c,0xc1,0x99,0x66,0x09,0x92,0x31,0x84,0x67,0x9a,0x25,0x48,0x06,0x2a,0x80,
+0x33,0x38,0xd0,0x00,0x99,0x63,0x18,0x83,0x34,0x98,0xc6,0x10,0x1a,0xc8,0xd6,0x80,
+0xe0,0x70,0x03,0x1b,0x54,0xc4,0x2c,0x83,0xb2,0xb4,0xc1,0x18,0x82,0x03,0xcd,0x12,
+0x30,0x63,0x08,0x0f,0x34,0x4b,0xc0,0x0c,0x54,0x00,0x6e,0xa0,0xbc,0xc1,0x32,0xc7,
+0xa0,0x06,0x70,0x00,0x61,0x1c,0x84,0x03,0x01,0x00,0x00,0x00,0x4e,0x00,0x00,0x00,
+0x76,0x52,0x4c,0xcc,0x73,0xd3,0x24,0x05,0x64,0xec,0xcd,0x8d,0xcc,0xe5,0x87,0x46,
+0xc6,0x50,0x8a,0x89,0x79,0xee,0xdb,0x54,0x8a,0x89,0x79,0xee,0xdd,0x1a,0x88,0x89,
+0x79,0x68,0x73,0x20,0x26,0xe6,0xa9,0xed,0x81,0x98,0x98,0xc7,0x36,0x0b,0x62,0x62,
+0x9e,0xdb,0x32,0x88,0x89,0x79,0x72,0xd3,0x20,0x26,0xe6,0xd9,0x8d,0x83,0x98,0x98,
+0xa7,0xb7,0x95,0x62,0x62,0x9e,0xbb,0x27,0x2d,0x20,0x63,0x6f,0x6e,0x64,0x2e,0x3a,
+0x34,0x35,0x56,0x62,0x08,0x4e,0x53,0xd9,0xba,0xb5,0x14,0x02,0xf3,0xe0,0xf5,0x25,
+0x2c,0x82,0xd3,0x0c,0xbe,0xe0,0x34,0xd3,0x8d,0x9b,0x88,0x21,0x38,0xcd,0x60,0xd7,
+0x24,0x01,0x63,0xec,0xcd,0x8d,0xcc,0x45,0x87,0x44,0x80,0x8c,0xbd,0xb9,0x91,0xb9,
+0xfc,0xc4,0xd0,0x90,0x02,0x8c,0xb1,0x37,0x37,0x32,0x97,0x1f,0x73,0x29,0x26,0xe6,
+0xc1,0x71,0x7b,0x29,0x26,0xe6,0xc1,0x77,0xfb,0x28,0x04,0xe6,0xa9,0x6f,0x52,0x01,
+0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x13,0x73,0x63,0x18,0x83,0xc0,0x3c,0xb6,0x41,0x08,
+0x4e,0x33,0x58,0x47,0x31,0x31,0x4f,0x5d,0x1f,0xc3,0x22,0x38,0xcd,0xe0,0x0b,0x4e,
+0x33,0xe1,0xbc,0xa5,0x18,0x82,0xd3,0x0c,0x77,0x6e,0x20,0xc5,0xc4,0x3c,0xb5,0x4e,
+0x3a,0x40,0xc6,0xde,0xdc,0xc8,0x5c,0x7e,0x64,0x70,0x2c,0xa4,0x98,0x98,0xa7,0xee,
+0x6f,0x20,0x11,0x9c,0x66,0xf0,0x05,0xa7,0x99,0xec,0x82,0x10,0x9c,0xa6,0x32,0x93,
+0x42,0x60,0x1e,0x7b,0xb7,0x98,0x62,0x62,0x9e,0xbc,0x36,0x16,0x43,0x70,0x9a,0x0a,
+0xa7,0x6d,0xa4,0x98,0x98,0xc7,0xbe,0x8d,0xa4,0x98,0x98,0xc7,0xce,0x0d,0xc6,0x10,
+0x9c,0x66,0xc0,0x7b,0x12,0x02,0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x33,0x13,0x73,0x06,
+0x8b,0xe0,0x34,0x83,0x2f,0x38,0xcd,0x64,0xd3,0x07,0x50,0x10,0xcd,0x14,0x61,0xe6,
+0x61,0x08,0x4e,0x53,0xd5,0x36,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x4a,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x07,0x00,0x00,0x00,
+0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0xb9,0x61,0x0c,0x04,0x10,
+0x1e,0xe1,0x19,0xc6,0x40,0x02,0xe1,0x11,0x1e,0x00,0x00,0x00,0x63,0x08,0xcd,0x63,
+0x15,0xc1,0x31,0x84,0x06,0xb2,0x8b,0xe0,0x18,0x42,0x13,0x59,0x46,0x70,0x0c,0xa1,
+0x71,0x6c,0x23,0x38,0x16,0x02,0x04,0xc7,0x64,0x61,0x1a,0x37,0x16,0x01,0x04,0x48,
+0x35,0xc7,0x20,0x79,0xcf,0x58,0x04,0x10,0x20,0xd5,0x1c,0xc3,0x07,0x06,0xd0,0x58,
+0x04,0x10,0x20,0xd5,0x1c,0x43,0x18,0x88,0x41,0x34,0x16,0x01,0x04,0x48,0x35,0xc7,
+0x30,0x06,0x64,0xe0,0x98,0x47,0xd0,0xc0,0x80,0xa0,0x89,0x01,0x41,0x23,0x03,0x82,
+0x63,0x21,0x40,0x70,0x50,0x66,0x70,0x06,0x68,0x90,0x06,0x58,0x06,0xe1,0x40,0x00,
+0x25,0x00,0x00,0x00,0x56,0x52,0x4c,0xcc,0x73,0xd3,0x56,0x41,0x4c,0xcc,0x53,0xdb,
+0x05,0x31,0x31,0xcf,0x6d,0x19,0xc4,0xc4,0x3c,0xba,0x6d,0x10,0x13,0xf3,0xf4,0xd6,
+0x41,0x08,0xc0,0xb2,0x18,0x46,0x21,0x38,0x4d,0x85,0x9b,0x46,0x21,0x38,0x4d,0xb5,
+0x9b,0x8a,0x21,0x00,0xcb,0x82,0xdf,0x66,0x62,0x08,0x4e,0x53,0xdd,0xb7,0x9d,0x18,
+0x82,0xd3,0x54,0xb7,0x6e,0x28,0x86,0xe0,0x34,0xd5,0xdd,0xdb,0x47,0x31,0x31,0x4f,
+0x9d,0x9b,0x87,0x21,0x00,0xcb,0x52,0xdf,0x06,0x62,0x08,0xc0,0xb2,0xd4,0xbc,0x59,
+0x10,0x82,0xd3,0x54,0x96,0x62,0x08,0x4e,0x53,0xe1,0xb6,0x85,0x14,0x13,0xf3,0xd8,
+0xb4,0x8d,0x14,0x13,0xf3,0xd8,0xb9,0x89,0x18,0x02,0xb0,0x2c,0xf6,0x6d,0x24,0x86,
+0x00,0x2c,0x8b,0xcd,0x1b,0x87,0x21,0x38,0x4d,0x55,0xd3,0xd6,0x30,0x54,0xc0,0x72,
+0x00,0x05,0xd1,0x4c,0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x19,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,
+0x24,0x4a,0x60,0x04,0x80,0xc2,0x0c,0x00,0x00,0x00,0x00,0x00,0x63,0x08,0xcd,0x33,
+0x16,0x01,0x04,0x48,0x34,0xc7,0x00,0x49,0xcf,0x58,0x04,0x10,0x28,0xd1,0x1c,0xc3,
+0x44,0x39,0x58,0x85,0x03,0x01,0x00,0x00,0x0a,0x00,0x00,0x00,0x26,0x41,0x08,0xc0,
+0xb2,0x18,0x45,0x21,0x00,0xcb,0xb2,0x5b,0x04,0x31,0x31,0x8f,0x6d,0x13,0xc4,0xc4,
+0x3c,0xb9,0x35,0x0c,0x15,0xb0,0x58,0x05,0x31,0x31,0x4f,0x7f,0x00,0x05,0xd1,0x4c,
+0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x1b,0x00,0x00,0x00,
+0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0xca,0x60,0x04,
+0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x00,0x63,0x08,0xcd,0x33,0x16,0x01,0x04,0xca,
+0x34,0xc7,0x20,0x51,0xcf,0x1c,0x43,0x45,0x41,0x73,0x0c,0x16,0x15,0xcd,0x31,0x5c,
+0x94,0x83,0x58,0x38,0x10,0x00,0x00,0x00,0x0b,0x00,0x00,0x00,0x76,0x51,0x4c,0xcc,
+0x53,0xdb,0x86,0x51,0x4c,0xcc,0x53,0xe7,0x36,0x41,0x4c,0xcc,0x63,0x5b,0x05,0x31,
+0x31,0x8f,0x6e,0x16,0xc4,0xc4,0x3c,0xbd,0x51,0x10,0x02,0xb0,0x2c,0xd6,0x30,0x54,
+0xc0,0x72,0x00,0x05,0xd1,0x4c,0x11,0x06,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x2c,0x00,0x00,0x00,0x13,0x04,0x45,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,
+0x24,0xca,0xa0,0x04,0x46,0x00,0x8a,0x80,0xc0,0x08,0x00,0x00,0x63,0x08,0x0d,0x34,
+0xdc,0x30,0x49,0xc4,0x2c,0x03,0x11,0x50,0x63,0x08,0xcd,0x33,0xdc,0x50,0x49,0xc4,
+0x2c,0x03,0x21,0x58,0x63,0x08,0x4d,0x34,0xdc,0x70,0x49,0xc4,0x2c,0x03,0x31,0x60,
+0x63,0x08,0x8d,0x33,0xdc,0x90,0x49,0x84,0x69,0x22,0x70,0xc3,0x27,0x1c,0x08,0x00,
+0x17,0x00,0x00,0x00,0x96,0x51,0x4c,0xcc,0x53,0xdf,0x66,0x41,0x08,0xcc,0x83,0xdb,
+0x04,0x31,0x31,0x4f,0x6d,0x15,0xc4,0xc4,0x3c,0xb7,0x61,0x10,0x02,0xf3,0xf0,0x76,
+0x41,0x4c,0xcc,0xb3,0x1f,0x81,0x11,0x11,0x13,0x15,0x35,0x37,0x90,0x2c,0x4e,0xf4,
+0x47,0x87,0x54,0xd7,0x17,0x70,0x2c,0x4e,0xf4,0x47,0x87,0x74,0x02,0xc8,0xe2,0x44,
+0x7f,0x74,0x48,0xb9,0x69,0x14,0x02,0xf3,0xd4,0xb8,0x6d,0x18,0x11,0x31,0x55,0xc0,
+0x62,0x0d,0x43,0x05,0x2c,0x07,0x50,0x10,0xcd,0x14,0x61,0x46,0x31,0x08,0xcc,0x03,
+0x00,0x00,0x00,0x00,0x71,0x20,0x00,0x00,0x12,0x00,0x00,0x00,0x66,0x40,0x54,0x82,
+0x23,0x19,0xc3,0xa0,0x20,0x8b,0x1d,0x18,0x4f,0x84,0x34,0x53,0x61,0x03,0xc4,0xe3,
+0x58,0x85,0x05,0x14,0xbe,0x34,0x45,0xb5,0x21,0x10,0x82,0x23,0x15,0x46,0x30,0x2c,
+0xc8,0x64,0x02,0x06,0xf0,0x3c,0x91,0x73,0x19,0x00,0xe1,0x4b,0x53,0x64,0x0a,0x84,
+0x84,0x34,0x85,0x25,0x0c,0x92,0x20,0x59,0xc1,0x20,0x30,0x8f,0x2d,0x10,0x95,0x84,
+0x34,0x00,0x00,0x00,0x00,0x00,0x00,0x00};
diff --git a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
index e64bfb1c6c..3a2f2878a3 100644
--- a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
+++ b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
@@ -46,6 +46,7 @@
 #include "tgsi/tgsi_dump.h"
 
 #include "util/u_memory.h"
+#include "util/u_math.h"
 
 #include <llvm/Module.h>
 #include <llvm/CallingConv.h>
@@ -157,8 +158,8 @@ void gallivm_cpu_jit_compile(struct gallivm_cpu_engine *cpu, struct gallivm_prog
    llvm::ExistingModuleProvider *mp = new llvm::ExistingModuleProvider(mod);
    llvm::ExecutionEngine *ee = cpu->engine;
    assert(ee);
-   /*FIXME : remove */
-   ee->DisableLazyCompilation();
+   /*FIXME : why was this disabled ? we need it for pow/sqrt/... */
+   ee->DisableLazyCompilation(false);
    ee->addModuleProvider(mp);
 
    llvm::Function *func = func_for_shader(prog);
@@ -201,7 +202,6 @@ int gallivm_cpu_vs_exec(struct gallivm_prog *prog,
    unsigned int i, j;
    unsigned slot;
    vertex_shader_runner runner = reinterpret_cast<vertex_shader_runner>(prog->function);
-
    assert(runner);
 
    for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
diff --git a/src/gallium/auxiliary/gallivm/instructions.cpp b/src/gallium/auxiliary/gallivm/instructions.cpp
index a82dc30306..599975d5ad 100644
--- a/src/gallium/auxiliary/gallivm/instructions.cpp
+++ b/src/gallium/auxiliary/gallivm/instructions.cpp
@@ -83,6 +83,7 @@ Instructions::Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicB
    m_llvmPow   = 0;
    m_llvmFloor = 0;
    m_llvmFlog  = 0;
+   m_llvmFexp  = 0;
    m_llvmLit  = 0;
    m_fmtPtr = 0;
 
@@ -92,194 +93,271 @@ Instructions::Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicB
    m_mod = ParseBitcodeFile(buffer);
 }
 
+llvm::BasicBlock * Instructions::currentBlock() const
+{
+   return m_builder.GetInsertBlock();
+}
+
+llvm::Value * Instructions::abs(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   Value *xabs  = callFAbs(vec[0]);
+   Value *yabs  = callFAbs(vec[1]);
+   Value *zabs  = callFAbs(vec[2]);
+   Value *wabs  = callFAbs(vec[3]);
+   return vectorFromVals(xabs, yabs, zabs, wabs);
+}
+
 llvm::Value * Instructions::add(llvm::Value *in1, llvm::Value *in2)
 {
    return m_builder.CreateAdd(in1, in2, name("add"));
 }
 
-llvm::Value * Instructions::madd(llvm::Value *in1, llvm::Value *in2,
-                                 llvm::Value *in3)
+llvm::Value * Instructions::arl(llvm::Value *in)
 {
-   Value *mulRes = mul(in1, in2);
-   return add(mulRes, in3);
+   return floor(in);
 }
- 
-llvm::Value * Instructions::mul(llvm::Value *in1, llvm::Value *in2)
+
+void Instructions::beginLoop()
 {
-   return m_builder.CreateMul(in1, in2, name("mul"));
+   BasicBlock *begin = BasicBlock::Create(name("loop"), m_func,0);
+   BasicBlock *end = BasicBlock::Create(name("endloop"), m_func,0);
+
+   m_builder.CreateBr(begin);
+   Loop loop;
+   loop.begin = begin;
+   loop.end   = end;
+   m_builder.SetInsertPoint(begin);
+   m_loopStack.push(loop);
 }
 
-const char * Instructions::name(const char *prefix)
+void Instructions::bgnSub(unsigned label)
 {
-   ++m_idx;
-   snprintf(m_name, 32, "%s%d", prefix, m_idx);
-   return m_name;
+   llvm::Function *func = findFunction(label);
+
+   Function::arg_iterator args = func->arg_begin();
+   Value *ptr_INPUT = args++;
+   ptr_INPUT->setName("INPUT");
+   m_storage->pushArguments(ptr_INPUT);
+
+   llvm::BasicBlock *entry = BasicBlock::Create("entry", func, 0);
+
+   m_func = func;
+   m_builder.SetInsertPoint(entry);
 }
 
-llvm::Value * Instructions::dp3(llvm::Value *in1, llvm::Value *in2)
+void Instructions::brk()
 {
-   Value *mulRes = mul(in1, in2);
-   Value *x = m_builder.CreateExtractElement(mulRes,
-                                                          m_storage->constantInt(0),
-                                                          name("extractx"));
-   Value *y = m_builder.CreateExtractElement(mulRes,
-                                                          m_storage->constantInt(1),
-                                                          name("extracty"));
-   Value *z = m_builder.CreateExtractElement(mulRes,
-                                                          m_storage->constantInt(2),
-                                                          name("extractz"));
-   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
-   Value *dot3 = m_builder.CreateAdd(xy, z, name("dot3"));
-   return vectorFromVals(dot3, dot3, dot3, dot3);
+   assert(!m_loopStack.empty());
+   BasicBlock *unr = BasicBlock::Create(name("unreachable"), m_func,0);
+   m_builder.CreateBr(m_loopStack.top().end);
+   m_builder.SetInsertPoint(unr);
 }
 
-llvm::Value *Instructions::callFSqrt(llvm::Value *val)
+void Instructions::cal(int label, llvm::Value *input)
 {
-   if (!m_llvmFSqrt) {
-      // predeclare the intrinsic
-      std::vector<const Type*> fsqrtArgs;
-      fsqrtArgs.push_back(Type::FloatTy);
-      PAListPtr fsqrtPal;
-      FunctionType* fsqrtType = FunctionType::get(
-         /*Result=*/Type::FloatTy,
-         /*Params=*/fsqrtArgs,
-         /*isVarArg=*/false);
-      m_llvmFSqrt = Function::Create(
-         /*Type=*/fsqrtType,
-         /*Linkage=*/GlobalValue::ExternalLinkage,
-         /*Name=*/"llvm.sqrt.f32", m_mod);
-      m_llvmFSqrt->setCallingConv(CallingConv::C);
-      m_llvmFSqrt->setParamAttrs(fsqrtPal);
-   }
-   CallInst *call = m_builder.CreateCall(m_llvmFSqrt, val,
-                                         name("sqrt"));
-   call->setCallingConv(CallingConv::C);
-   call->setTailCall(false);
-   return call;
+   std::vector<Value*> params;
+   params.push_back(input);
+   llvm::Function *func = findFunction(label);
+
+   m_builder.CreateCall(func, params.begin(), params.end());
 }
 
-llvm::Value * Instructions::rsq(llvm::Value *in1)
+llvm::Value * Instructions::ceil(llvm::Value *in)
 {
-   Value *x = m_builder.CreateExtractElement(in1,
-                                             m_storage->constantInt(0),
-                                             name("extractx"));
-   Value *abs  = callFAbs(x);
-   Value *sqrt = callFSqrt(abs);
-
-   Value *rsqrt = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)),
-                                       sqrt,
-                                       name("rsqrt"));
-   return vectorFromVals(rsqrt, rsqrt, rsqrt, rsqrt);
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callCeil(vec[0]), callCeil(vec[1]),
+                         callCeil(vec[2]), callCeil(vec[3]));
 }
 
-llvm::Value * Instructions::vectorFromVals(llvm::Value *x, llvm::Value *y,
-                                           llvm::Value *z, llvm::Value *w)
+llvm::Value * Instructions::clamp(llvm::Value *in1)
 {
-   Constant *const_vec = Constant::getNullValue(m_floatVecType);
-   Value *res = m_builder.CreateInsertElement(const_vec, x,
-                                              m_storage->constantInt(0),
-                                              name("vecx"));
-   res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1),
-                               name("vecxy"));
-   res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2),
-                               name("vecxyz"));
-   if (w)
-      res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3),
-                                          name("vecxyzw"));
-   return res;
+   llvm::Value *zero = constVector(0.0f, 0.0f, 0.0f, 0.0f);
+   llvm::Value *one = constVector(1.0f, 1.0f, 1.0f, 1.0f);
+   return min( max(zero, in1), one);
 }
 
-llvm::Value *Instructions::callFAbs(llvm::Value *val)
+llvm::Value * Instructions::cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
 {
-   if (!m_llvmFAbs) {
-      // predeclare the intrinsic
-      std::vector<const Type*> fabsArgs;
-      fabsArgs.push_back(Type::FloatTy);
-      PAListPtr fabsPal;
-      FunctionType* fabsType = FunctionType::get(
-         /*Result=*/Type::FloatTy,
-         /*Params=*/fabsArgs,
-         /*isVarArg=*/false);
-      m_llvmFAbs = Function::Create(
-         /*Type=*/fabsType,
-         /*Linkage=*/GlobalValue::ExternalLinkage,
-         /*Name=*/"fabs", m_mod);
-      m_llvmFAbs->setCallingConv(CallingConv::C);
-      m_llvmFAbs->setParamAttrs(fabsPal);
-   }
-   CallInst *call = m_builder.CreateCall(m_llvmFAbs, val,
-                                         name("fabs"));
-   call->setCallingConv(CallingConv::C);
+   llvm::Function *func = m_mod->getFunction("cmp");
+   assert(func);
+
+   std::vector<Value*> params;
+   params.push_back(in1);
+   params.push_back(in2);
+   params.push_back(in3);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end(), name("cmpres"));
    call->setTailCall(false);
    return call;
 }
 
-llvm::Value * Instructions::lit(llvm::Value *in)
+llvm::Value * Instructions::cnd(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
 {
-   if (!m_llvmLit) {
-      m_llvmLit = m_mod->getFunction("lit");
-   }
-   CallInst *call = m_builder.CreateCall(m_llvmLit, in, name("litres"));
-   call->setCallingConv(CallingConv::C);
-   call->setTailCall(false);
-   return call;
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   std::vector<llvm::Value*> vec3 = extractVector(in3);
+   Constant *half = ConstantFP::get(APFloat(0.5f));
+
+   Value *xcmp  = m_builder.CreateFCmpOGT(vec1[0], half, name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec2[0], vec3[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOGT(vec1[1], half, name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec2[1], vec3[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOGT(vec1[2], half, name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec2[2], vec3[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOGT(vec1[3], half, name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec2[3], vec3[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
 }
 
-llvm::Value * Instructions::sub(llvm::Value *in1, llvm::Value *in2)
+llvm::Value * Instructions::cnd0(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
 {
-   Value *res = m_builder.CreateSub(in1, in2, name("sub"));
-   return res;
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   std::vector<llvm::Value*> vec3 = extractVector(in3);
+   Constant *zero = Constant::getNullValue(Type::FloatTy);
+
+   Value *xcmp  = m_builder.CreateFCmpOGE(vec1[0], zero, name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec2[0], vec3[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOGE(vec1[1], zero, name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec2[1], vec3[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOGE(vec1[2], zero, name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec2[2], vec3[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOGE(vec1[3], zero, name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec2[3], vec3[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
 }
 
-llvm::Value * Instructions::callPow(llvm::Value *val1, llvm::Value *val2)
+llvm::Value * Instructions::cos(llvm::Value *in)
 {
-   if (!m_llvmPow) {
-      // predeclare the intrinsic
-      std::vector<const Type*> powArgs;
-      powArgs.push_back(Type::FloatTy);
-      powArgs.push_back(Type::FloatTy);
-      PAListPtr powPal;
-      FunctionType* powType = FunctionType::get(
-         /*Result=*/Type::FloatTy,
-         /*Params=*/powArgs,
-         /*isVarArg=*/false);
-      m_llvmPow = Function::Create(
-         /*Type=*/powType,
-         /*Linkage=*/GlobalValue::ExternalLinkage,
-         /*Name=*/"llvm.pow.f32", m_mod);
-      m_llvmPow->setCallingConv(CallingConv::C);
-      m_llvmPow->setParamAttrs(powPal);
-   }
-   std::vector<Value*> params;
-   params.push_back(val1);
-   params.push_back(val2);
-   CallInst *call = m_builder.CreateCall(m_llvmPow, params.begin(), params.end(),
-                                         name("pow"));
-   call->setCallingConv(CallingConv::C);
+#if 0
+   llvm::Function *func = m_mod->getFunction("vcos");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("cosres"));
    call->setTailCall(false);
    return call;
+#else
+   std::vector<llvm::Value*> elems = extractVector(in);
+   Function *func = m_mod->getFunction("cosf");
+   assert(func);
+   CallInst *cos = m_builder.CreateCall(func, elems[0], name("cosres"));
+   cos->setCallingConv(CallingConv::C);
+   cos->setTailCall(true);
+   return vectorFromVals(cos, cos, cos, cos);
+#endif
 }
 
-llvm::Value * Instructions::pow(llvm::Value *in1, llvm::Value *in2)
+llvm::Value * Instructions::cross(llvm::Value *in1, llvm::Value *in2)
 {
    Value *x1 = m_builder.CreateExtractElement(in1,
                                               m_storage->constantInt(0),
                                               name("x1"));
+   Value *y1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(1),
+                                              name("y1"));
+   Value *z1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(2),
+                                              name("z1"));
+
    Value *x2 = m_builder.CreateExtractElement(in2,
                                               m_storage->constantInt(0),
                                               name("x2"));
-   llvm::Value *val = callPow(x1, x2);
-   return vectorFromVals(val, val, val, val);
+   Value *y2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(1),
+                                              name("y2"));
+   Value *z2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(2),
+                                              name("z2"));
+   Value *y1z2 = mul(y1, z2);
+   Value *z1y2 = mul(z1, y2);
+
+   Value *z1x2 = mul(z1, x2);
+   Value *x1z2 = mul(x1, z2);
+
+   Value *x1y2 = mul(x1, y2);
+   Value *y1x2 = mul(y1, x2);
+
+   return vectorFromVals(sub(y1z2, z1y2), sub(z1x2, x1z2), sub(x1y2, y1x2));
 }
 
-llvm::Value * Instructions::rcp(llvm::Value *in1)
+llvm::Value * Instructions::ddx(llvm::Value *in)
 {
-   Value *x1 = m_builder.CreateExtractElement(in1,
-                                              m_storage->constantInt(0),
-                                              name("x1"));
-   Value *res = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)),
-                                     x1, name("rcp"));
-   return vectorFromVals(res, res, res, res);
+   // FIXME
+   assert(0);
+}
+
+llvm::Value * Instructions::ddy(llvm::Value *in)
+{
+   // FIXME
+   assert(0);
+}
+
+llvm::Value * Instructions::div(llvm::Value *in1, llvm::Value *in2)
+{
+   return m_builder.CreateFDiv(in1, in2, name("div"));
+}
+
+llvm::Value * Instructions::dot2add(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
+{
+   Value *mulRes = mul(in1, in2);
+   Value *x = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(0),
+                                                          name("extractx"));
+   Value *y = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(1),
+                                                          name("extracty"));
+   Value *z = m_builder.CreateExtractElement(in3,
+                                                          m_storage->constantInt(2),
+                                                          name("extractz"));
+   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
+   Value *dot2add = m_builder.CreateAdd(xy, z, name("dot2add"));
+   return vectorFromVals(dot2add, dot2add, dot2add, dot2add);
+}
+
+llvm::Value * Instructions::dp2(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   Value *x = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(0),
+                                                          name("extractx"));
+   Value *y = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(1),
+                                                          name("extracty"));
+   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
+   return vectorFromVals(xy, xy, xy, xy);
+}
+
+llvm::Value * Instructions::dp3(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   Value *x = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(0),
+                                                          name("extractx"));
+   Value *y = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(1),
+                                                          name("extracty"));
+   Value *z = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(2),
+                                                          name("extractz"));
+   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
+   Value *dot3 = m_builder.CreateAdd(xy, z, name("dot3"));
+   return vectorFromVals(dot3, dot3, dot3, dot3);
 }
 
 llvm::Value * Instructions::dp4(llvm::Value *in1, llvm::Value *in2)
@@ -321,6 +399,53 @@ llvm::Value * Instructions::dst(llvm::Value *in1, llvm::Value *in2)
                          ry, z, w);
 }
 
+void Instructions::elseop()
+{
+   assert(!m_ifStack.empty());
+   BasicBlock *ifend = BasicBlock::Create(name("ifend"), m_func,0);
+   m_builder.CreateBr(ifend);
+   m_builder.SetInsertPoint(m_ifStack.top());
+   currentBlock()->setName(name("ifelse"));
+   m_ifStack.pop();
+   m_ifStack.push(ifend);
+}
+
+void Instructions::endif()
+{
+   assert(!m_ifStack.empty());
+   m_builder.CreateBr(m_ifStack.top());
+   m_builder.SetInsertPoint(m_ifStack.top());
+   m_ifStack.pop();
+}
+
+void Instructions::endLoop()
+{
+   assert(!m_loopStack.empty());
+   Loop loop = m_loopStack.top();
+   m_builder.CreateBr(loop.begin);
+   loop.end->moveAfter(currentBlock());
+   m_builder.SetInsertPoint(loop.end);
+   m_loopStack.pop();
+}
+
+void Instructions::end()
+{
+   m_builder.CreateRetVoid();
+}
+
+void Instructions::endSub()
+{
+   m_func = 0;
+   m_builder.SetInsertPoint(0);
+}
+
+llvm::Value * Instructions::exp(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callFExp(vec[0]), callFExp(vec[1]),
+                             callFExp(vec[2]), callFExp(vec[3]));
+}
+
 llvm::Value * Instructions::ex2(llvm::Value *in)
 {
    llvm::Value *val = callPow(ConstantFP::get(APFloat(2.f)),
@@ -330,31 +455,6 @@ llvm::Value * Instructions::ex2(llvm::Value *in)
    return vectorFromVals(val, val, val, val);
 }
 
-llvm::Value * Instructions::callFloor(llvm::Value *val)
-{
-   if (!m_llvmFloor) {
-      // predeclare the intrinsic
-      std::vector<const Type*> floorArgs;
-      floorArgs.push_back(Type::FloatTy);
-      PAListPtr floorPal;
-      FunctionType* floorType = FunctionType::get(
-         /*Result=*/Type::FloatTy,
-         /*Params=*/floorArgs,
-         /*isVarArg=*/false);
-      m_llvmFloor = Function::Create(
-         /*Type=*/floorType,
-         /*Linkage=*/GlobalValue::ExternalLinkage,
-         /*Name=*/"floorf", m_mod);
-      m_llvmFloor->setCallingConv(CallingConv::C);
-      m_llvmFloor->setParamAttrs(floorPal);
-   }
-   CallInst *call =  m_builder.CreateCall(m_llvmFloor, val,
-                                          name("floorf"));
-   call->setCallingConv(CallingConv::C);
-   call->setTailCall(false);
-   return call;
-}
-
 llvm::Value * Instructions::floor(llvm::Value *in)
 {
    std::vector<llvm::Value*> vec = extractVector(in);
@@ -362,42 +462,52 @@ llvm::Value * Instructions::floor(llvm::Value *in)
                          callFloor(vec[2]), callFloor(vec[3]));
 }
 
-llvm::Value * Instructions::arl(llvm::Value *in)
-{
-   return floor(in);
-}
-
 llvm::Value * Instructions::frc(llvm::Value *in)
 {
    llvm::Value *flr = floor(in);
    return sub(in, flr);
 }
 
-llvm::Value * Instructions::callFLog(llvm::Value *val)
+void Instructions::ifop(llvm::Value *in)
 {
-   if (!m_llvmFlog) {
-      // predeclare the intrinsic
-      std::vector<const Type*> flogArgs;
-      flogArgs.push_back(Type::FloatTy);
-      PAListPtr flogPal;
-      FunctionType* flogType = FunctionType::get(
-         /*Result=*/Type::FloatTy,
-         /*Params=*/flogArgs,
-         /*isVarArg=*/false);
-      m_llvmFlog = Function::Create(
-         /*Type=*/flogType,
-         /*Linkage=*/GlobalValue::ExternalLinkage,
-         /*Name=*/"logf", m_mod);
-      m_llvmFlog->setCallingConv(CallingConv::C);
-      m_llvmFlog->setParamAttrs(flogPal);
-   }
-   CallInst *call = m_builder.CreateCall(m_llvmFlog, val,
-                                         name("logf"));
-   call->setCallingConv(CallingConv::C);
+   BasicBlock *ifthen = BasicBlock::Create(name("ifthen"), m_func,0);
+   BasicBlock *ifend = BasicBlock::Create(name("ifthenend"), m_func,0);
+
+   //BasicBlock *yblock = new BasicBlock(name("yblock"), m_func,0);
+   //BasicBlock *zblock = new BasicBlock(name("zblock"), m_func,0);
+   //BasicBlock *wblock = new BasicBlock(name("wblock"), m_func,0);
+
+   Constant *float0 = Constant::getNullValue(Type::FloatTy);
+
+   Value *x = m_builder.CreateExtractElement(in, m_storage->constantInt(0),
+                                             name("extractx"));
+   Value *xcmp = m_builder.CreateFCmpUNE(x, float0, name("xcmp"));
+   m_builder.CreateCondBr(xcmp, ifthen, ifend);
+   //m_builder.SetInsertPoint(yblock);
+
+   m_builder.SetInsertPoint(ifthen);
+   m_ifStack.push(ifend);
+}
+
+llvm::Value * Instructions::kil(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("kil");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("kilpres"));
    call->setTailCall(false);
    return call;
 }
 
+llvm::Value * Instructions::lerp(llvm::Value *in1, llvm::Value *in2,
+                                 llvm::Value *in3)
+{
+   llvm::Value *m = mul(in1, in2);
+   llvm::Value *vec1 = constVector(1.f, 1.f, 1.f, 1.f);
+   llvm::Value *s = sub(vec1, in1);
+   return add(m, mul(s, in3));
+}
+
 llvm::Value * Instructions::lg2(llvm::Value *in)
 {
    std::vector<llvm::Value*> vec = extractVector(in);
@@ -407,142 +517,176 @@ llvm::Value * Instructions::lg2(llvm::Value *in)
                              callFLog(vec[2]), callFLog(vec[3])), const_vec);
 }
 
-llvm::Value * Instructions::min(llvm::Value *in1, llvm::Value *in2)
+llvm::Value * Instructions::lit(llvm::Value *in)
+{
+   if (!m_llvmLit) {
+      m_llvmLit = m_mod->getFunction("lit");
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmLit, in, name("litres"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::log(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callFLog(vec[0]), callFLog(vec[1]),
+                             callFLog(vec[2]), callFLog(vec[3]));
+}
+
+llvm::Value * Instructions::madd(llvm::Value *in1, llvm::Value *in2,
+                                 llvm::Value *in3)
+{
+   Value *mulRes = mul(in1, in2);
+   return add(mulRes, in3);
+}
+
+llvm::Value * Instructions::max(llvm::Value *in1, llvm::Value *in2)
 {
    std::vector<llvm::Value*> vec1 = extractVector(in1);
    std::vector<llvm::Value*> vec2 = extractVector(in2);
 
-   Value *xcmp  = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
+   Value *xcmp  = m_builder.CreateFCmpOGT(vec1[0], vec2[0],
+                                          name("xcmp"));
    Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0],
                                         name("selx"));
 
-   Value *ycmp  = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
+   Value *ycmp  = m_builder.CreateFCmpOGT(vec1[1], vec2[1],
+                                          name("ycmp"));
    Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1],
                                         name("sely"));
 
-   Value *zcmp  = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
+   Value *zcmp  = m_builder.CreateFCmpOGT(vec1[2], vec2[2],
+                                          name("zcmp"));
    Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2],
                                         name("selz"));
 
-   Value *wcmp  = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
+   Value *wcmp  = m_builder.CreateFCmpOGT(vec1[3], vec2[3],
+                                          name("wcmp"));
    Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3],
                                         name("selw"));
 
    return vectorFromVals(selx, sely, selz, selw);
 }
 
-llvm::Value * Instructions::max(llvm::Value *in1, llvm::Value *in2)
+llvm::Value * Instructions::min(llvm::Value *in1, llvm::Value *in2)
 {
    std::vector<llvm::Value*> vec1 = extractVector(in1);
    std::vector<llvm::Value*> vec2 = extractVector(in2);
 
-   Value *xcmp  = m_builder.CreateFCmpOGT(vec1[0], vec2[0],
-                                          name("xcmp"));
+   Value *xcmp  = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
    Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0],
                                         name("selx"));
 
-   Value *ycmp  = m_builder.CreateFCmpOGT(vec1[1], vec2[1],
-                                          name("ycmp"));
+   Value *ycmp  = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
    Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1],
                                         name("sely"));
 
-   Value *zcmp  = m_builder.CreateFCmpOGT(vec1[2], vec2[2],
-                                          name("zcmp"));
+   Value *zcmp  = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
    Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2],
                                         name("selz"));
 
-   Value *wcmp  = m_builder.CreateFCmpOGT(vec1[3], vec2[3],
-                                          name("wcmp"));
+   Value *wcmp  = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
    Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3],
                                         name("selw"));
 
    return vectorFromVals(selx, sely, selz, selw);
 }
 
-void Instructions::printVector(llvm::Value *val)
+llvm::Value * Instructions::mul(llvm::Value *in1, llvm::Value *in2)
 {
-   static const char *frmt = "Vector is [%f, %f, %f, %f]\x0A";
+   return m_builder.CreateMul(in1, in2, name("mul"));
+}
 
-   if (!m_fmtPtr) {
-      Constant *format = ConstantArray::get(frmt, true);
-      ArrayType *arrayTy = ArrayType::get(IntegerType::get(8), strlen(frmt) + 1);
-      GlobalVariable* globalFormat = new GlobalVariable(
-         /*Type=*/arrayTy,
-         /*isConstant=*/true,
-         /*Linkage=*/GlobalValue::InternalLinkage,
-         /*Initializer=*/0, // has initializer, specified below
-         /*Name=*/name(".str"),
-         m_mod);
-      globalFormat->setInitializer(format);
+llvm::Value * Instructions::neg(llvm::Value *in)
+{
+   Value *neg = m_builder.CreateNeg(in, name("neg"));
+   return neg;
+}
 
-      Constant* const_int0 = Constant::getNullValue(IntegerType::get(32));
-      std::vector<Constant*> const_ptr_21_indices;
-      const_ptr_21_indices.push_back(const_int0);
-      const_ptr_21_indices.push_back(const_int0);
-      m_fmtPtr = ConstantExpr::getGetElementPtr(globalFormat,
-                                                &const_ptr_21_indices[0], const_ptr_21_indices.size());
-   }
+llvm::Value * Instructions::nrm(llvm::Value *in)
+{
+   llvm::Value *v = rsq(in);
+   return mul(v, in);
+}
 
-   Function *func_printf = m_mod->getFunction("printf");
-   if (!func_printf)
-      func_printf = declarePrintf();
-   assert(func_printf);
-   std::vector<llvm::Value*> vec = extractVector(val);
-   Value *dx = m_builder.CreateFPExt(vec[0], Type::DoubleTy, name("dx"));
-   Value *dy = m_builder.CreateFPExt(vec[1], Type::DoubleTy, name("dy"));
-   Value *dz = m_builder.CreateFPExt(vec[2], Type::DoubleTy, name("dz"));
-   Value *dw = m_builder.CreateFPExt(vec[3], Type::DoubleTy, name("dw"));
-   std::vector<Value*> params;
-   params.push_back(m_fmtPtr);
-   params.push_back(dx);
-   params.push_back(dy);
-   params.push_back(dz);
-   params.push_back(dw);
-   CallInst *call = m_builder.CreateCall(func_printf, params.begin(), params.end(),
-                                         name("printf"));
-   call->setCallingConv(CallingConv::C);
-   call->setTailCall(true);
+llvm::Value * Instructions::pow(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *x2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(0),
+                                              name("x2"));
+   llvm::Value *val = callPow(x1, x2);
+   return vectorFromVals(val, val, val, val);
 }
 
-llvm::Function * Instructions::declarePrintf()
+llvm::Value * Instructions::rcp(llvm::Value *in1)
 {
-   std::vector<const Type*> args;
-   PAListPtr params;
-   FunctionType* funcTy = FunctionType::get(
-      /*Result=*/IntegerType::get(32),
-      /*Params=*/args,
-      /*isVarArg=*/true);
-   Function* func_printf = Function::Create(
-      /*Type=*/funcTy,
-      /*Linkage=*/GlobalValue::ExternalLinkage,
-      /*Name=*/"printf", m_mod);
-   func_printf->setCallingConv(CallingConv::C);
-   func_printf->setParamAttrs(params);
-   return func_printf;
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *res = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)),
+                                     x1, name("rcp"));
+   return vectorFromVals(res, res, res, res);
+}
+
+llvm::Value * Instructions::rsq(llvm::Value *in1)
+{
+   Value *x = m_builder.CreateExtractElement(in1,
+                                             m_storage->constantInt(0),
+                                             name("extractx"));
+   Value *abs  = callFAbs(x);
+   Value *sqrt = callFSqrt(abs);
+
+   Value *rsqrt = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)),
+                                       sqrt,
+                                       name("rsqrt"));
+   return vectorFromVals(rsqrt, rsqrt, rsqrt, rsqrt);
 }
 
+llvm::Value * Instructions::scs(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("scs");
+   assert(func);
 
-llvm::Value * Instructions::sgt(llvm::Value *in1, llvm::Value *in2)
+   CallInst *call = m_builder.CreateCall(func, in, name("scsres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::seq(llvm::Value *in1, llvm::Value *in2)
 {
    Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
    Constant *const0f = Constant::getNullValue(Type::FloatTy);
 
    std::vector<llvm::Value*> vec1 = extractVector(in1);
    std::vector<llvm::Value*> vec2 = extractVector(in2);
-   Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], name("xcmp"));
+
+   Value *xcmp = m_builder.CreateFCmpOEQ(vec1[0], vec2[0], name("xcmp"));
    Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
 
-   Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], name("ycmp"));
+   Value *ycmp = m_builder.CreateFCmpOEQ(vec1[1], vec2[1], name("ycmp"));
    Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
 
-   Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], name("zcmp"));
+   Value *zcmp = m_builder.CreateFCmpOEQ(vec1[2], vec2[2], name("zcmp"));
    Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
 
-   Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], name("wcmp"));
+   Value *wcmp = m_builder.CreateFCmpOEQ(vec1[3], vec2[3], name("wcmp"));
    Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
 
    return vectorFromVals(x, y, z, w);
 }
+
+llvm::Value * Instructions::sfl(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   return vectorFromVals(const0f, const0f, const0f, const0f);
+}
+
 llvm::Value * Instructions::sge(llvm::Value *in1, llvm::Value *in2)
 {
    Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
@@ -566,157 +710,118 @@ llvm::Value * Instructions::sge(llvm::Value *in1, llvm::Value *in2)
    return vectorFromVals(x, y, z, w);
 }
 
-
-llvm::Value * Instructions::slt(llvm::Value *in1, llvm::Value *in2)
+llvm::Value * Instructions::sgt(llvm::Value *in1, llvm::Value *in2)
 {
    Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
    Constant *const0f = Constant::getNullValue(Type::FloatTy);
 
    std::vector<llvm::Value*> vec1 = extractVector(in1);
    std::vector<llvm::Value*> vec2 = extractVector(in2);
-
-   Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
+   Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], name("xcmp"));
    Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
 
-   Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
+   Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], name("ycmp"));
    Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
 
-   Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
+   Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], name("zcmp"));
    Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
 
-   Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
+   Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], name("wcmp"));
    Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
 
    return vectorFromVals(x, y, z, w);
 }
 
-llvm::Value * Instructions::cross(llvm::Value *in1, llvm::Value *in2)
+llvm::Value * Instructions::sin(llvm::Value *in)
 {
-   Value *x1 = m_builder.CreateExtractElement(in1,
-                                              m_storage->constantInt(0),
-                                              name("x1"));
-   Value *y1 = m_builder.CreateExtractElement(in1,
-                                              m_storage->constantInt(1),
-                                              name("y1"));
-   Value *z1 = m_builder.CreateExtractElement(in1,
-                                              m_storage->constantInt(2),
-                                              name("z1"));
+   llvm::Function *func = m_mod->getFunction("vsin");
+   assert(func);
 
-   Value *x2 = m_builder.CreateExtractElement(in2,
-                                              m_storage->constantInt(0),
-                                              name("x2"));
-   Value *y2 = m_builder.CreateExtractElement(in2,
-                                              m_storage->constantInt(1),
-                                              name("y2"));
-   Value *z2 = m_builder.CreateExtractElement(in2,
-                                              m_storage->constantInt(2),
-                                              name("z2"));
-   Value *y1z2 = mul(y1, z2);
-   Value *z1y2 = mul(z1, y2);
+   CallInst *call = m_builder.CreateCall(func, in, name("sinres"));
+   call->setTailCall(false);
+   return call;
+}
 
-   Value *z1x2 = mul(z1, x2);
-   Value *x1z2 = mul(x1, z2);
+llvm::Value * Instructions::sle(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
 
-   Value *x1y2 = mul(x1, y2);
-   Value *y1x2 = mul(y1, x2);
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
 
-   return vectorFromVals(sub(y1z2, z1y2), sub(z1x2, x1z2), sub(x1y2, y1x2));
-}
+   Value *xcmp = m_builder.CreateFCmpOLE(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
 
+   Value *ycmp = m_builder.CreateFCmpOLE(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
 
-llvm::Value * Instructions::abs(llvm::Value *in)
-{
-   std::vector<llvm::Value*> vec = extractVector(in);
-   Value *xabs  = callFAbs(vec[0]);
-   Value *yabs  = callFAbs(vec[1]);
-   Value *zabs  = callFAbs(vec[2]);
-   Value *wabs  = callFAbs(vec[3]);
-   return vectorFromVals(xabs, yabs, zabs, wabs);
+   Value *zcmp = m_builder.CreateFCmpOLE(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOLE(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
 }
 
-void Instructions::ifop(llvm::Value *in)
+llvm::Value * Instructions::slt(llvm::Value *in1, llvm::Value *in2)
 {
-   BasicBlock *ifthen = BasicBlock::Create(name("ifthen"), m_func,0);
-   BasicBlock *ifend = BasicBlock::Create(name("ifthenend"), m_func,0);
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
 
-   //BasicBlock *yblock = new BasicBlock(name("yblock"), m_func,0);
-   //BasicBlock *zblock = new BasicBlock(name("zblock"), m_func,0);
-   //BasicBlock *wblock = new BasicBlock(name("wblock"), m_func,0);
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
 
-   Constant *float0 = Constant::getNullValue(Type::FloatTy);
+   Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
 
-   Value *x = m_builder.CreateExtractElement(in, m_storage->constantInt(0),
-                                             name("extractx"));
-   Value *xcmp = m_builder.CreateFCmpUNE(x, float0, name("xcmp"));
-   m_builder.CreateCondBr(xcmp, ifthen, ifend);
-   //m_builder.SetInsertPoint(yblock);
+   Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
 
-   m_builder.SetInsertPoint(ifthen);
-   m_ifStack.push(ifend);
-}
+   Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
 
-llvm::BasicBlock * Instructions::currentBlock() const
-{
-   return m_builder.GetInsertBlock();
-}
+   Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
 
-void Instructions::elseop()
-{
-   assert(!m_ifStack.empty());
-   BasicBlock *ifend = BasicBlock::Create(name("ifend"), m_func,0);
-   m_builder.CreateBr(ifend);
-   m_builder.SetInsertPoint(m_ifStack.top());
-   currentBlock()->setName(name("ifelse"));
-   m_ifStack.pop();
-   m_ifStack.push(ifend);
+   return vectorFromVals(x, y, z, w);
 }
 
-void Instructions::endif()
+llvm::Value * Instructions::sne(llvm::Value *in1, llvm::Value *in2)
 {
-   assert(!m_ifStack.empty());
-   m_builder.CreateBr(m_ifStack.top());
-   m_builder.SetInsertPoint(m_ifStack.top());
-   m_ifStack.pop();
-}
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
 
-llvm::Value * Instructions::lerp(llvm::Value *in1, llvm::Value *in2,
-                                 llvm::Value *in3)
-{
-   llvm::Value *m = mul(in1, in2);
-   llvm::Value *vec1 = constVector(1.f, 1.f, 1.f, 1.f);
-   llvm::Value *s = sub(vec1, in1);
-   return add(m, mul(s, in3));
-}
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
 
-void Instructions::beginLoop()
-{
-   BasicBlock *begin = BasicBlock::Create(name("loop"), m_func,0);
-   BasicBlock *end = BasicBlock::Create(name("endloop"), m_func,0);
+   Value *xcmp = m_builder.CreateFCmpONE(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
 
-   m_builder.CreateBr(begin);
-   Loop loop;
-   loop.begin = begin;
-   loop.end   = end;
-   m_builder.SetInsertPoint(begin);
-   m_loopStack.push(loop);
+   Value *ycmp = m_builder.CreateFCmpONE(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpONE(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpONE(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
 }
 
-void Instructions::endLoop()
+llvm::Value * Instructions::str(llvm::Value *in1, llvm::Value *in2)
 {
-   assert(!m_loopStack.empty());
-   Loop loop = m_loopStack.top();
-   m_builder.CreateBr(loop.begin);
-   loop.end->moveAfter(currentBlock());
-   m_builder.SetInsertPoint(loop.end);
-   m_loopStack.pop();
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+
+   return vectorFromVals(const1f, const1f, const1f, const1f);
 }
 
-void Instructions::brk()
+llvm::Value * Instructions::sub(llvm::Value *in1, llvm::Value *in2)
 {
-   assert(!m_loopStack.empty());
-   BasicBlock *unr = BasicBlock::Create(name("unreachable"), m_func,0);
-   m_builder.CreateBr(m_loopStack.top().end);
-   m_builder.SetInsertPoint(unr);
+   Value *res = m_builder.CreateSub(in1, in2, name("sub"));
+   return res;
 }
 
 llvm::Value * Instructions::trunc(llvm::Value *in)
@@ -741,18 +846,298 @@ llvm::Value * Instructions::trunc(llvm::Value *in)
    return vectorFromVals(fx, fy, fz, fw);
 }
 
-void Instructions::end()
+llvm::Value * Instructions::x2d(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
 {
-   m_builder.CreateRetVoid();
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   std::vector<llvm::Value*> vec3 = extractVector(in3);
+
+   Value *x2x3 = m_builder.CreateMul( vec2[0], vec3[0], name("x2x3"));
+   Value *y2y3 = m_builder.CreateMul( vec2[1], vec3[1], name("y2y3"));
+   Value *x1px2x3 = m_builder.CreateAdd (vec1[0], x2x3, name("x1 + x2x3"));
+   Value *x1px2x3py2y3 = m_builder.CreateAdd (x1px2x3, y2y3, name("x1 + x2x3 + y2y3"));
+
+   Value *x2z3 = m_builder.CreateMul( vec2[0], vec3[2], name("x2z3"));
+   Value *y2w3 = m_builder.CreateMul( vec2[1], vec3[3], name("y2w3"));
+   Value *y1px2z3 = m_builder.CreateAdd (vec1[1], x2z3, name("y1 + x2z3"));
+   Value *y1px2z3py2w3 = m_builder.CreateAdd (y1px2z3, y2w3, name("y1 + x2z3 + y2w3"));
+
+   return vectorFromVals(x1px2x3py2y3, y1px2z3py2w3, x1px2x3py2y3, y1px2z3py2w3);
 }
 
-void Instructions::cal(int label, llvm::Value *input)
+void Instructions::printVector(llvm::Value *val)
 {
+   static const char *frmt = "Vector is [%f, %f, %f, %f]\x0A";
+
+   if (!m_fmtPtr) {
+      Constant *format = ConstantArray::get(frmt, true);
+      ArrayType *arrayTy = ArrayType::get(IntegerType::get(8), strlen(frmt) + 1);
+      GlobalVariable* globalFormat = new GlobalVariable(
+         /*Type=*/arrayTy,
+         /*isConstant=*/true,
+         /*Linkage=*/GlobalValue::InternalLinkage,
+         /*Initializer=*/0, // has initializer, specified below
+         /*Name=*/name(".str"),
+         m_mod);
+      globalFormat->setInitializer(format);
+
+      Constant* const_int0 = Constant::getNullValue(IntegerType::get(32));
+      std::vector<Constant*> const_ptr_21_indices;
+      const_ptr_21_indices.push_back(const_int0);
+      const_ptr_21_indices.push_back(const_int0);
+      m_fmtPtr = ConstantExpr::getGetElementPtr(globalFormat,
+                                                &const_ptr_21_indices[0], const_ptr_21_indices.size());
+   }
+
+   Function *func_printf = m_mod->getFunction("printf");
+   if (!func_printf)
+      func_printf = declarePrintf();
+   assert(func_printf);
+   std::vector<llvm::Value*> vec = extractVector(val);
+   Value *dx = m_builder.CreateFPExt(vec[0], Type::DoubleTy, name("dx"));
+   Value *dy = m_builder.CreateFPExt(vec[1], Type::DoubleTy, name("dy"));
+   Value *dz = m_builder.CreateFPExt(vec[2], Type::DoubleTy, name("dz"));
+   Value *dw = m_builder.CreateFPExt(vec[3], Type::DoubleTy, name("dw"));
    std::vector<Value*> params;
-   params.push_back(input);
-   llvm::Function *func = findFunction(label);
+   params.push_back(m_fmtPtr);
+   params.push_back(dx);
+   params.push_back(dy);
+   params.push_back(dz);
+   params.push_back(dw);
+   CallInst *call = m_builder.CreateCall(func_printf, params.begin(), params.end(),
+                                         name("printf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(true);
+}
 
-   m_builder.CreateCall(func, params.begin(), params.end());
+const char * Instructions::name(const char *prefix)
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::Value * Instructions::callCeil(llvm::Value *val)
+{
+   if (!m_llvmCeil) {
+      // predeclare the intrinsic
+      std::vector<const Type*> ceilArgs;
+      ceilArgs.push_back(Type::FloatTy);
+      AttrListPtr ceilPal;
+      FunctionType* ceilType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/ceilArgs,
+         /*isVarArg=*/false);
+      m_llvmCeil = Function::Create(
+         /*Type=*/ceilType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"ceilf", m_mod);
+      m_llvmCeil->setCallingConv(CallingConv::C);
+      m_llvmCeil->setAttributes(ceilPal);
+   }
+   CallInst *call =  m_builder.CreateCall(m_llvmCeil, val,
+                                          name("ceilf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value *Instructions::callFAbs(llvm::Value *val)
+{
+   if (!m_llvmFAbs) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fabsArgs;
+      fabsArgs.push_back(Type::FloatTy);
+      AttrListPtr fabsPal;
+      FunctionType* fabsType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fabsArgs,
+         /*isVarArg=*/false);
+      m_llvmFAbs = Function::Create(
+         /*Type=*/fabsType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"fabs", m_mod);
+      m_llvmFAbs->setCallingConv(CallingConv::C);
+      m_llvmFAbs->setAttributes(fabsPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFAbs, val,
+                                         name("fabs"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callFExp(llvm::Value *val)
+{
+   if (!m_llvmFexp) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fexpArgs;
+      fexpArgs.push_back(Type::FloatTy);
+      AttrListPtr fexpPal;
+      FunctionType* fexpType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fexpArgs,
+         /*isVarArg=*/false);
+      m_llvmFexp = Function::Create(
+         /*Type=*/fexpType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"expf", m_mod);
+      m_llvmFexp->setCallingConv(CallingConv::C);
+      m_llvmFexp->setAttributes(fexpPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFexp, val,
+                                         name("expf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callFLog(llvm::Value *val)
+{
+   if (!m_llvmFlog) {
+      // predeclare the intrinsic
+      std::vector<const Type*> flogArgs;
+      flogArgs.push_back(Type::FloatTy);
+      AttrListPtr flogPal;
+      FunctionType* flogType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/flogArgs,
+         /*isVarArg=*/false);
+      m_llvmFlog = Function::Create(
+         /*Type=*/flogType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"logf", m_mod);
+      m_llvmFlog->setCallingConv(CallingConv::C);
+      m_llvmFlog->setAttributes(flogPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFlog, val,
+                                         name("logf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callFloor(llvm::Value *val)
+{
+   if (!m_llvmFloor) {
+      // predeclare the intrinsic
+      std::vector<const Type*> floorArgs;
+      floorArgs.push_back(Type::FloatTy);
+      AttrListPtr floorPal;
+      FunctionType* floorType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/floorArgs,
+         /*isVarArg=*/false);
+      m_llvmFloor = Function::Create(
+         /*Type=*/floorType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"floorf", m_mod);
+      m_llvmFloor->setCallingConv(CallingConv::C);
+      m_llvmFloor->setAttributes(floorPal);
+   }
+   CallInst *call =  m_builder.CreateCall(m_llvmFloor, val,
+                                          name("floorf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value *Instructions::callFSqrt(llvm::Value *val)
+{
+   if (!m_llvmFSqrt) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fsqrtArgs;
+      fsqrtArgs.push_back(Type::FloatTy);
+      AttrListPtr fsqrtPal;
+      FunctionType* fsqrtType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fsqrtArgs,
+         /*isVarArg=*/false);
+      m_llvmFSqrt = Function::Create(
+         /*Type=*/fsqrtType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"llvm.sqrt.f32", m_mod);
+      m_llvmFSqrt->setCallingConv(CallingConv::C);
+      m_llvmFSqrt->setAttributes(fsqrtPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFSqrt, val,
+                                         name("sqrt"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callPow(llvm::Value *val1, llvm::Value *val2)
+{
+   if (!m_llvmPow) {
+      // predeclare the intrinsic
+      std::vector<const Type*> powArgs;
+      powArgs.push_back(Type::FloatTy);
+      powArgs.push_back(Type::FloatTy);
+      AttrListPtr powPal;
+      FunctionType* powType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/powArgs,
+         /*isVarArg=*/false);
+      m_llvmPow = Function::Create(
+         /*Type=*/powType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"llvm.pow.f32", m_mod);
+      m_llvmPow->setCallingConv(CallingConv::C);
+      m_llvmPow->setAttributes(powPal);
+   }
+   std::vector<Value*> params;
+   params.push_back(val1);
+   params.push_back(val2);
+   CallInst *call = m_builder.CreateCall(m_llvmPow, params.begin(), params.end(),
+                                         name("pow"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::vectorFromVals(llvm::Value *x, llvm::Value *y,
+                                           llvm::Value *z, llvm::Value *w)
+{
+   Constant *const_vec = Constant::getNullValue(m_floatVecType);
+   Value *res = m_builder.CreateInsertElement(const_vec, x,
+                                              m_storage->constantInt(0),
+                                              name("vecx"));
+   res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1),
+                               name("vecxy"));
+   res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2),
+                               name("vecxyz"));
+   if (w)
+      res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3),
+                                          name("vecxyzw"));
+   return res;
+}
+
+llvm::Value * Instructions::constVector(float x, float y, float z, float w)
+{
+   std::vector<Constant*> vec(4);
+   vec[0] = ConstantFP::get(APFloat(x));
+   vec[1] = ConstantFP::get(APFloat(y));
+   vec[2] = ConstantFP::get(APFloat(z));
+   vec[3] = ConstantFP::get(APFloat(w));
+   return ConstantVector::get(m_floatVecType, vec);
+}
+
+llvm::Function * Instructions::declarePrintf()
+{
+   std::vector<const Type*> args;
+   AttrListPtr params;
+   FunctionType* funcTy = FunctionType::get(
+      /*Result=*/IntegerType::get(32),
+      /*Params=*/args,
+      /*isVarArg=*/true);
+   Function* func_printf = Function::Create(
+      /*Type=*/funcTy,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Name=*/"printf", m_mod);
+   func_printf->setCallingConv(CallingConv::C);
+   func_printf->setAttributes(params);
+   return func_printf;
 }
 
 llvm::Function * Instructions::declareFunc(int label)
@@ -763,7 +1148,7 @@ llvm::Function * Instructions::declareFunc(int label)
    args.push_back(vecPtr);
    args.push_back(vecPtr);
    args.push_back(vecPtr);
-   PAListPtr params;
+   AttrListPtr params;
    FunctionType *funcType = FunctionType::get(
       /*Result=*/Type::VoidTy,
       /*Params=*/args,
@@ -774,31 +1159,10 @@ llvm::Function * Instructions::declareFunc(int label)
       /*Linkage=*/GlobalValue::ExternalLinkage,
       /*Name=*/name.c_str(), m_mod);
    func->setCallingConv(CallingConv::C);
-   func->setParamAttrs(params);
+   func->setAttributes(params);
    return func;
 }
 
-void Instructions::bgnSub(unsigned label)
-{
-   llvm::Function *func = findFunction(label);
-
-   Function::arg_iterator args = func->arg_begin();
-   Value *ptr_INPUT = args++;
-   ptr_INPUT->setName("INPUT");
-   m_storage->pushArguments(ptr_INPUT);
-
-   llvm::BasicBlock *entry = BasicBlock::Create("entry", func, 0);
-
-   m_func = func;
-   m_builder.SetInsertPoint(entry);
-}
-
-void Instructions::endSub()
-{
-   m_func = 0;
-   m_builder.SetInsertPoint(0);
-}
-
 llvm::Function * Instructions::findFunction(int label)
 {
    llvm::Function *func = m_functions[label];
@@ -809,17 +1173,6 @@ llvm::Function * Instructions::findFunction(int label)
    return func;
 }
 
-llvm::Value * Instructions::constVector(float x, float y, float z, float w)
-{
-   std::vector<Constant*> vec(4);
-   vec[0] = ConstantFP::get(APFloat(x));
-   vec[1] = ConstantFP::get(APFloat(y));
-   vec[2] = ConstantFP::get(APFloat(z));
-   vec[3] = ConstantFP::get(APFloat(w));
-   return ConstantVector::get(m_floatVecType, vec);
-}
-
-
 std::vector<llvm::Value*> Instructions::extractVector(llvm::Value *vec)
 {
    std::vector<llvm::Value*> elems(4);
@@ -834,69 +1187,7 @@ std::vector<llvm::Value*> Instructions::extractVector(llvm::Value *vec)
    return elems;
 }
 
-llvm::Value * Instructions::cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
-{
-   llvm::Function *func = m_mod->getFunction("cmp");
-   assert(func);
-
-   std::vector<Value*> params;
-   params.push_back(in1);
-   params.push_back(in2);
-   params.push_back(in3);
-   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end(), name("cmpres"));
-   call->setTailCall(false);
-   return call;
-}
-
-llvm::Value * Instructions::cos(llvm::Value *in)
-{
-#if 0
-   llvm::Function *func = m_mod->getFunction("vcos");
-   assert(func);
 
-   CallInst *call = m_builder.CreateCall(func, in, name("cosres"));
-   call->setTailCall(false);
-   return call;
-#else
-   std::vector<llvm::Value*> elems = extractVector(in);
-   Function *func = m_mod->getFunction("cosf");
-   assert(func);
-   CallInst *cos = m_builder.CreateCall(func, elems[0], name("cosres"));
-   cos->setCallingConv(CallingConv::C);
-   cos->setTailCall(true);
-   return vectorFromVals(cos, cos, cos, cos);
-#endif
-}
-
-llvm::Value * Instructions::scs(llvm::Value *in)
-{
-   llvm::Function *func = m_mod->getFunction("scs");
-   assert(func);
-
-   CallInst *call = m_builder.CreateCall(func, in, name("scsres"));
-   call->setTailCall(false);
-   return call;
-}
-
-llvm::Value * Instructions::kil(llvm::Value *in)
-{
-   llvm::Function *func = m_mod->getFunction("kil");
-   assert(func);
-
-   CallInst *call = m_builder.CreateCall(func, in, name("kilpres"));
-   call->setTailCall(false);
-   return call;
-}
-
-llvm::Value * Instructions::sin(llvm::Value *in)
-{
-   llvm::Function *func = m_mod->getFunction("vsin");
-   assert(func);
-
-   CallInst *call = m_builder.CreateCall(func, in, name("sinres"));
-   call->setTailCall(false);
-   return call;
-}
 #endif //MESA_LLVM
 
 
diff --git a/src/gallium/auxiliary/gallivm/instructions.h b/src/gallium/auxiliary/gallivm/instructions.h
index d286ce80c7..e18571251e 100644
--- a/src/gallium/auxiliary/gallivm/instructions.h
+++ b/src/gallium/auxiliary/gallivm/instructions.h
@@ -57,15 +57,24 @@ public:
    llvm::BasicBlock *currentBlock() const;
 
    llvm::Value *abs(llvm::Value *in1);
-   llvm::Value *arl(llvm::Value *in1);
    llvm::Value *add(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *arl(llvm::Value *in1);
    void         beginLoop();
    void         bgnSub(unsigned);
    void         brk();
    void         cal(int label, llvm::Value *input);
+   llvm::Value *ceil(llvm::Value *in);
+   llvm::Value *clamp(llvm::Value *in);
    llvm::Value *cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *cnd(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *cnd0(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
    llvm::Value *cos(llvm::Value *in);
    llvm::Value *cross(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *ddx(llvm::Value *in);
+   llvm::Value *ddy(llvm::Value *in);
+   llvm::Value *div(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dot2add(llvm::Value *in, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *dp2(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *dp3(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *dp4(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *dph(llvm::Value *in1, llvm::Value *in2);
@@ -75,6 +84,7 @@ public:
    void         endLoop();
    void         end();
    void         endSub();
+   llvm::Value *exp(llvm::Value *in);
    llvm::Value *ex2(llvm::Value *in);
    llvm::Value *floor(llvm::Value *in);
    llvm::Value *frc(llvm::Value *in);
@@ -82,32 +92,43 @@ public:
    llvm::Value *kil(llvm::Value *in);
    llvm::Value *lerp(llvm::Value *in1, llvm::Value *in2,
                      llvm::Value *in3);
-   llvm::Value *lit(llvm::Value *in);
    llvm::Value *lg2(llvm::Value *in);
+   llvm::Value *lit(llvm::Value *in);
+   llvm::Value *log(llvm::Value *in);
    llvm::Value *madd(llvm::Value *in1, llvm::Value *in2,
                      llvm::Value *in3);
-   llvm::Value *min(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *max(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *min(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *mul(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *neg(llvm::Value *in);
+   llvm::Value *nrm(llvm::Value *in);
    llvm::Value *pow(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *rcp(llvm::Value *in);
    llvm::Value *rsq(llvm::Value *in);
    llvm::Value *scs(llvm::Value *in);
+   llvm::Value *seq(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sfl(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *sge(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *sgt(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *sin(llvm::Value *in);
+   llvm::Value *sle(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *slt(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sne(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *str(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *sub(llvm::Value *in1, llvm::Value *in2);
    llvm::Value *trunc(llvm::Value *in);
+   llvm::Value *x2d(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
 
    void printVector(llvm::Value *val);
 private:
    const char *name(const char *prefix);
 
+   llvm::Value *callCeil(llvm::Value *val);
    llvm::Value *callFAbs(llvm::Value *val);
+   llvm::Value *callFExp(llvm::Value *val);
+   llvm::Value *callFLog(llvm::Value *val);
    llvm::Value *callFloor(llvm::Value *val);
    llvm::Value *callFSqrt(llvm::Value *val);
-   llvm::Value *callFLog(llvm::Value *val);
    llvm::Value *callPow(llvm::Value *val1, llvm::Value *val2);
 
    llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
@@ -125,16 +146,18 @@ private:
    llvm::Module             *m_mod;
    llvm::Function           *m_func;
    char                      m_name[32];
-   llvm::IRBuilder           m_builder;
+   llvm::IRBuilder<>         m_builder;
    int                       m_idx;
 
    llvm::VectorType *m_floatVecType;
 
+   llvm::Function   *m_llvmCeil;
    llvm::Function   *m_llvmFSqrt;
    llvm::Function   *m_llvmFAbs;
    llvm::Function   *m_llvmPow;
    llvm::Function   *m_llvmFloor;
    llvm::Function   *m_llvmFlog;
+   llvm::Function   *m_llvmFexp;
    llvm::Function   *m_llvmLit;
 
    llvm::Constant   *m_fmtPtr;
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
index efddc04e81..d5600fd22d 100644
--- a/src/gallium/auxiliary/gallivm/instructionssoa.cpp
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
@@ -90,68 +90,11 @@ llvm::Value * InstructionsSoa::vectorFromVals(llvm::Value *x, llvm::Value *y,
    return res;
 }
 
-std::vector<llvm::Value*> InstructionsSoa::arl(const std::vector<llvm::Value*> in)
-{
-   std::vector<llvm::Value*> res(4);
-
-   //Extract x's
-   llvm::Value *x1 = m_builder.CreateExtractElement(in[0],
-                                                    m_storage->constantInt(0),
-                                                    name("extractX"));
-   //cast it to an unsigned int
-   x1 = m_builder.CreateFPToUI(x1, IntegerType::get(32), name("x1IntCast"));
-
-   res[0] = x1;//vectorFromVals(x1, x2, x3, x4);
-   //only x is valid. the others shouldn't be necessary
-   /*
-   res[1] = Constant::getNullValue(m_floatVecType);
-   res[2] = Constant::getNullValue(m_floatVecType);
-   res[3] = Constant::getNullValue(m_floatVecType);
-   */
-
-   return res;
-}
-
-
-std::vector<llvm::Value*> InstructionsSoa::add(const std::vector<llvm::Value*> in1,
-                                               const std::vector<llvm::Value*> in2)
-{
-   std::vector<llvm::Value*> res(4);
-
-   res[0] = m_builder.CreateAdd(in1[0], in2[0], name("addx"));
-   res[1] = m_builder.CreateAdd(in1[1], in2[1], name("addy"));
-   res[2] = m_builder.CreateAdd(in1[2], in2[2], name("addz"));
-   res[3] = m_builder.CreateAdd(in1[3], in2[3], name("addw"));
-
-   return res;
-}
-
-std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> in1,
-                                               const std::vector<llvm::Value*> in2)
-{
-   std::vector<llvm::Value*> res(4);
-
-   res[0] = m_builder.CreateMul(in1[0], in2[0], name("mulx"));
-   res[1] = m_builder.CreateMul(in1[1], in2[1], name("muly"));
-   res[2] = m_builder.CreateMul(in1[2], in2[2], name("mulz"));
-   res[3] = m_builder.CreateMul(in1[3], in2[3], name("mulw"));
-
-   return res;
-}
-
 void InstructionsSoa::end()
 {
    m_builder.CreateRetVoid();
 }
 
-std::vector<llvm::Value*> InstructionsSoa::madd(const std::vector<llvm::Value*> in1,
-                                                const std::vector<llvm::Value*> in2,
-                                                const std::vector<llvm::Value*> in3)
-{
-   std::vector<llvm::Value*> res = mul(in1, in2);
-   return add(res, in3);
-}
-
 std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector)
 {
    std::vector<llvm::Value*> res(4);
@@ -171,6 +114,11 @@ std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector)
    return res;
 }
 
+llvm::IRBuilder<>* InstructionsSoa::getIRBuilder()
+{
+   return &m_builder;
+}
+
 void InstructionsSoa::createFunctionMap()
 {
    m_functionsMap[TGSI_OPCODE_ABS]   = "abs";
@@ -181,6 +129,7 @@ void InstructionsSoa::createFunctionMap()
    m_functionsMap[TGSI_OPCODE_POWER] = "pow";
    m_functionsMap[TGSI_OPCODE_LIT]   = "lit";
    m_functionsMap[TGSI_OPCODE_RSQ]   = "rsq";
+   m_functionsMap[TGSI_OPCODE_SLT]   = "slt";
 }
 
 void InstructionsSoa::createDependencies()
@@ -273,6 +222,41 @@ std::vector<llvm::Value*> InstructionsSoa::abs(const std::vector<llvm::Value*> i
    return callBuiltin(func, in1);
 }
 
+std::vector<llvm::Value*> InstructionsSoa::add(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateAdd(in1[0], in2[0], name("addx"));
+   res[1] = m_builder.CreateAdd(in1[1], in2[1], name("addy"));
+   res[2] = m_builder.CreateAdd(in1[2], in2[2], name("addz"));
+   res[3] = m_builder.CreateAdd(in1[3], in2[3], name("addw"));
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::arl(const std::vector<llvm::Value*> in)
+{
+   std::vector<llvm::Value*> res(4);
+
+   //Extract x's
+   llvm::Value *x1 = m_builder.CreateExtractElement(in[0],
+                                                    m_storage->constantInt(0),
+                                                    name("extractX"));
+   //cast it to an unsigned int
+   x1 = m_builder.CreateFPToUI(x1, IntegerType::get(32), name("x1IntCast"));
+
+   res[0] = x1;//vectorFromVals(x1, x2, x3, x4);
+   //only x is valid. the others shouldn't be necessary
+   /*
+   res[1] = Constant::getNullValue(m_floatVecType);
+   res[2] = Constant::getNullValue(m_floatVecType);
+   res[3] = Constant::getNullValue(m_floatVecType);
+   */
+
+   return res;
+}
+
 std::vector<llvm::Value*> InstructionsSoa::dp3(const std::vector<llvm::Value*> in1,
                                                const std::vector<llvm::Value*> in2)
 {
@@ -280,6 +264,98 @@ std::vector<llvm::Value*> InstructionsSoa::dp3(const std::vector<llvm::Value*> i
    return callBuiltin(func, in1, in2);
 }
 
+std::vector<llvm::Value*> InstructionsSoa::lit(const std::vector<llvm::Value*> in)
+{
+   llvm::Function *func = function(TGSI_OPCODE_LIT);
+   return callBuiltin(func, in);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::madd(const std::vector<llvm::Value*> in1,
+                                                const std::vector<llvm::Value*> in2,
+                                                const std::vector<llvm::Value*> in3)
+{
+   std::vector<llvm::Value*> res = mul(in1, in2);
+   return add(res, in3);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::max(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_MAX);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::min(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_MIN);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateMul(in1[0], in2[0], name("mulx"));
+   res[1] = m_builder.CreateMul(in1[1], in2[1], name("muly"));
+   res[2] = m_builder.CreateMul(in1[2], in2[2], name("mulz"));
+   res[3] = m_builder.CreateMul(in1[3], in2[3], name("mulw"));
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::pow(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_POWER);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::rsq(const std::vector<llvm::Value*> in)
+{
+   llvm::Function *func = function(TGSI_OPCODE_RSQ);
+   return callBuiltin(func, in);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::slt(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_SLT);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::sub(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateSub(in1[0], in2[0], name("subx"));
+   res[1] = m_builder.CreateSub(in1[1], in2[1], name("suby"));
+   res[2] = m_builder.CreateSub(in1[2], in2[2], name("subz"));
+   res[3] = m_builder.CreateSub(in1[3], in2[3], name("subw"));
+
+   return res;
+}
+
+void checkFunction(Function *func)
+{
+   for (Function::const_iterator BI = func->begin(), BE = func->end();
+        BI != BE; ++BI) {
+      const BasicBlock &BB = *BI;
+      for (BasicBlock::const_iterator II = BB.begin(), IE = BB.end();
+           II != IE; ++II) {
+         const Instruction &I = *II;
+         std::cout<< "Instr = "<<I;
+         for (unsigned op = 0, E = I.getNumOperands(); op != E; ++op) {
+            const Value *Op = I.getOperand(op);
+            std::cout<< "\top = "<<Op<<"("<<op<<")"<<std::endl;
+            //I->setOperand(op, V);
+  }
+      }
+   }
+}
+
 llvm::Value * InstructionsSoa::allocaTemp()
 {
    VectorType *vector   = VectorType::get(Type::FloatTy, 4);
@@ -399,46 +475,6 @@ std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std
    return allocaToResult(allocaPtr);
 }
 
-std::vector<llvm::Value*> InstructionsSoa::pow(const std::vector<llvm::Value*> in1,
-                                               const std::vector<llvm::Value*> in2)
-{
-   llvm::Function *func = function(TGSI_OPCODE_POWER);
-   return callBuiltin(func, in1, in2);
-}
-
-std::vector<llvm::Value*> InstructionsSoa::min(const std::vector<llvm::Value*> in1,
-                                               const std::vector<llvm::Value*> in2)
-{
-   llvm::Function *func = function(TGSI_OPCODE_MIN);
-   return callBuiltin(func, in1, in2);
-}
-
-
-std::vector<llvm::Value*> InstructionsSoa::max(const std::vector<llvm::Value*> in1,
-                                               const std::vector<llvm::Value*> in2)
-{
-   llvm::Function *func = function(TGSI_OPCODE_MAX);
-   return callBuiltin(func, in1, in2);
-}
-
-void checkFunction(Function *func)
-{
-   for (Function::const_iterator BI = func->begin(), BE = func->end();
-        BI != BE; ++BI) {
-      const BasicBlock &BB = *BI;
-      for (BasicBlock::const_iterator II = BB.begin(), IE = BB.end();
-           II != IE; ++II) {
-         const Instruction &I = *II;
-         std::cout<< "Instr = "<<I;
-         for (unsigned op = 0, E = I.getNumOperands(); op != E; ++op) {
-            const Value *Op = I.getOperand(op);
-            std::cout<< "\top = "<<Op<<"("<<op<<")"<<std::endl;
-            //I->setOperand(op, V);
-  }
-      }
-   }
-}
-
 void InstructionsSoa::injectFunction(llvm::Function *originalFunc, int op)
 {
    assert(originalFunc);
@@ -458,8 +494,8 @@ void InstructionsSoa::injectFunction(llvm::Function *originalFunc, int op)
       func = Function::Create(originalFunc->getFunctionType(), GlobalValue::ExternalLinkage,
                               originalFunc->getName(), currentModule());
       func->setCallingConv(CallingConv::C);
-      const PAListPtr pal;
-      func->setParamAttrs(pal);
+      const AttrListPtr pal;
+      func->setAttributes(pal);
       currentModule()->dump();
    } else {
       DenseMap<const Value*, Value *> val;
@@ -483,28 +519,4 @@ void InstructionsSoa::injectFunction(llvm::Function *originalFunc, int op)
    }
 }
 
-std::vector<llvm::Value*> InstructionsSoa::sub(const std::vector<llvm::Value*> in1,
-                                               const std::vector<llvm::Value*> in2)
-{
-   std::vector<llvm::Value*> res(4);
-
-   res[0] = m_builder.CreateSub(in1[0], in2[0], name("subx"));
-   res[1] = m_builder.CreateSub(in1[1], in2[1], name("suby"));
-   res[2] = m_builder.CreateSub(in1[2], in2[2], name("subz"));
-   res[3] = m_builder.CreateSub(in1[3], in2[3], name("subw"));
-
-   return res;
-}
-
-std::vector<llvm::Value*> InstructionsSoa::lit(const std::vector<llvm::Value*> in)
-{
-   llvm::Function *func = function(TGSI_OPCODE_LIT);
-   return callBuiltin(func, in);
-}
-
-std::vector<llvm::Value*> InstructionsSoa::rsq(const std::vector<llvm::Value*> in)
-{
-   llvm::Function *func = function(TGSI_OPCODE_RSQ);
-   return callBuiltin(func, in);
-}
 
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.h b/src/gallium/auxiliary/gallivm/instructionssoa.h
index 3e20b652dd..d6831e0a6b 100644
--- a/src/gallium/auxiliary/gallivm/instructionssoa.h
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.h
@@ -69,11 +69,14 @@ public:
    std::vector<llvm::Value*> pow(const std::vector<llvm::Value*> in1,
                                  const std::vector<llvm::Value*> in2);
    std::vector<llvm::Value*> rsq(const std::vector<llvm::Value*> in1);
+   std::vector<llvm::Value*> slt(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
    std::vector<llvm::Value*> sub(const std::vector<llvm::Value*> in1,
                                  const std::vector<llvm::Value*> in2);
    void         end();
 
    std::vector<llvm::Value*> extractVector(llvm::Value *vector);
+   llvm::IRBuilder<>*  getIRBuilder();
 private:
    const char * name(const char *prefix) const;
    llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
@@ -96,7 +99,7 @@ private:
                                          const std::vector<llvm::Value*> in3);
    void injectFunction(llvm::Function *originalFunc, int op = TGSI_OPCODE_LAST);
 private:
-   llvm::IRBuilder  m_builder;
+   llvm::IRBuilder<>  m_builder;
    StorageSoa *m_storage;
 
    std::map<int, std::string> m_functionsMap;
diff --git a/src/gallium/auxiliary/gallivm/soabuiltins.c b/src/gallium/auxiliary/gallivm/soabuiltins.c
index 78f84510e2..cb85e1734e 100644
--- a/src/gallium/auxiliary/gallivm/soabuiltins.c
+++ b/src/gallium/auxiliary/gallivm/soabuiltins.c
@@ -36,6 +36,8 @@ typedef __attribute__(( ext_vector_type(4) )) float float4;
 
 extern float fabsf(float val);
 
+/* helpers */
+
 float4 absvec(float4 vec)
 {
    float4 res;
@@ -47,6 +49,58 @@ float4 absvec(float4 vec)
    return res;
 }
 
+float4 maxvec(float4 a, float4 b)
+{
+   return (float4){(a.x > b.x) ? a.x : b.x,
+         (a.y > b.y) ? a.y : b.y,
+         (a.z > b.z) ? a.z : b.z,
+         (a.w > b.w) ? a.w : b.w};
+}
+
+float4 minvec(float4 a, float4 b)
+{
+   return (float4){(a.x < b.x) ? a.x : b.x,
+         (a.y < b.y) ? a.y : b.y,
+         (a.z < b.z) ? a.z : b.z,
+         (a.w < b.w) ? a.w : b.w};
+}
+
+extern float powf(float num, float p);
+extern float sqrtf(float x);
+
+float4 powvec(float4 vec, float4 q)
+{
+   float4 p;
+   p.x = powf(vec.x, q.x);
+   p.y = powf(vec.y, q.y);
+   p.z = powf(vec.z, q.z);
+   p.w = powf(vec.w, q.w);
+   return p;
+}
+
+float4 sqrtvec(float4 vec)
+{
+   float4 p;
+   p.x = sqrtf(vec.x);
+   p.y = sqrtf(vec.y);
+   p.z = sqrtf(vec.z);
+   p.w = sqrtf(vec.w);
+   return p;
+}
+
+float4 sltvec(float4 v1, float4 v2)
+{
+   float4 p;
+   p.x = (v1.x < v2.x) ? 1.0 : 0.0;
+   p.y = (v1.y < v2.y) ? 1.0 : 0.0;
+   p.z = (v1.z < v2.z) ? 1.0 : 0.0;
+   p.w = (v1.w < v2.w) ? 1.0 : 0.0;
+   return p;
+}
+
+
+/* instructions */
+
 void abs(float4 *res,
          float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w)
 {
@@ -69,7 +123,6 @@ void dp3(float4 *res,
    res[3] = dot;
 }
 
-
 void dp4(float4 *res,
          float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
          float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
@@ -83,35 +136,25 @@ void dp4(float4 *res,
    res[3] = dot;
 }
 
-extern float powf(float num, float p);
-extern float sqrtf(float x);
-
-float4 powvec(float4 vec, float4 q)
-{
-   float4 p;
-   p.x = powf(vec.x, q.x);
-   p.y = powf(vec.y, q.y);
-   p.z = powf(vec.z, q.z);
-   p.w = powf(vec.w, q.w);
-   return p;
-}
-
-void pow(float4 *res,
-         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
-         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+void lit(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w)
 {
-   res[0] = powvec(tmp0x, tmp1x);
-   res[1] = res[0];
-   res[2] = res[0];
-   res[3] = res[0];
-}
+   const float4 zerovec = (float4) {0.0, 0.0, 0.0, 0.0};
+   const float4 min128 = (float4) {-128.f, -128.f, -128.f, -128.f};
+   const float4 plus128 = (float4) {128.f,  128.f,  128.f,  128.f};
 
-float4 minvec(float4 a, float4 b)
-{
-   return (float4){(a.x < b.x) ? a.x : b.x,
-         (a.y < b.y) ? a.y : b.y,
-         (a.z < b.z) ? a.z : b.z,
-         (a.w < b.w) ? a.w : b.w};
+   res[0] = (float4){1.0, 1.0, 1.0, 1.0};
+   if (tmp0x.x > 0) {
+      float4 tmpy = maxvec(tmp0y, zerovec);
+      float4 tmpw = minvec(tmp0w, plus128);
+      tmpw = maxvec(tmpw, min128);
+      res[1] = tmp0x;
+      res[2] = powvec(tmpy, tmpw);
+   } else {
+      res[1] = zerovec;
+      res[2] = zerovec;
+   }
+   res[3] = (float4){1.0, 1.0, 1.0, 1.0};
 }
 
 void min(float4 *res,
@@ -125,14 +168,6 @@ void min(float4 *res,
 }
 
 
-float4 maxvec(float4 a, float4 b)
-{
-   return (float4){(a.x > b.x) ? a.x : b.x,
-         (a.y > b.y) ? a.y : b.y,
-         (a.z > b.z) ? a.z : b.z,
-         (a.w > b.w) ? a.w : b.w};
-}
-
 void max(float4 *res,
          float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
          float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
@@ -143,37 +178,14 @@ void max(float4 *res,
    res[3] = maxvec(tmp0w, tmp1w);
 }
 
-
-void lit(float4 *res,
-         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w)
-{
-   const float4 zerovec = (float4) {0.0, 0.0, 0.0, 0.0};
-   const float4 min128 = (float4) {-128.f, -128.f, -128.f, -128.f};
-   const float4 plus128 = (float4) {128.f,  128.f,  128.f,  128.f};
-
-   res[0] = (float4){1.0, 1.0, 1.0, 1.0};
-   if (tmp0x.x > 0) {
-      float4 tmpy = maxvec(tmp0y, zerovec);
-      float4 tmpw = minvec(tmp0w, plus128);
-      tmpw = maxvec(tmpw, min128);
-      res[1] = tmp0x;
-      res[2] = powvec(tmpy, tmpw);
-   } else {
-      res[1] = zerovec;
-      res[2] = zerovec;
-   }
-   res[3] = (float4){1.0, 1.0, 1.0, 1.0};
-}
-
-
-float4 sqrtvec(float4 vec)
+void pow(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
 {
-   float4 p;
-   p.x = sqrtf(vec.x);
-   p.y = sqrtf(vec.y);
-   p.z = sqrtf(vec.z);
-   p.w = sqrtf(vec.w);
-   return p;
+   res[0] = powvec(tmp0x, tmp1x);
+   res[1] = res[0];
+   res[2] = res[0];
+   res[3] = res[0];
 }
 
 void rsq(float4 *res,
@@ -185,3 +197,14 @@ void rsq(float4 *res,
    res[2] = onevec/sqrtvec(absvec(tmp0z));
    res[3] = onevec/sqrtvec(absvec(tmp0w));
 }
+
+void slt(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   res[0] = sltvec(tmp0x, tmp1x);
+   res[1] = sltvec(tmp0y, tmp1y);
+   res[2] = sltvec(tmp0z, tmp1z);
+   res[3] = sltvec(tmp0w, tmp1w);
+}
+
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.cpp b/src/gallium/auxiliary/gallivm/storagesoa.cpp
index 78d754371f..4fc075cf6d 100644
--- a/src/gallium/auxiliary/gallivm/storagesoa.cpp
+++ b/src/gallium/auxiliary/gallivm/storagesoa.cpp
@@ -93,7 +93,7 @@ void StorageSoa::declareImmediates()
       std::vector<float> vals(4);
       std::vector<Constant*> channelArray;
 
-      vals[0] = vec[0]; vals[1] = vec[0]; vals[2] = vec[0]; vals[3] = vec[0];
+      vals[0] = vec[0]; vals[1] = vec[1]; vals[2] = vec[2]; vals[3] = vec[3];
       llvm::Constant *xChannel = createConstGlobalVector(vals);
 
       vals[0] = vec[1]; vals[1] = vec[1]; vals[2] = vec[1]; vals[3] = vec[1];
@@ -144,22 +144,43 @@ std::vector<llvm::Value*> StorageSoa::inputElement(llvm::Value *idx)
    return res;
 }
 
-std::vector<llvm::Value*> StorageSoa::constElement(llvm::Value *idx)
+llvm::Value* StorageSoa::unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value* vector, int cc)
 {
-   std::vector<llvm::Value*> res(4);
+   std::vector<llvm::Value*> x(4);
+   x[0] = m_builder->CreateExtractElement(vector,
+                                           constantInt(cc),
+                                           name("x"));
+
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   Constant *constVector = Constant::getNullValue(vectorType);
+   Value *res = m_builder->CreateInsertElement(constVector, x[0],
+                                              constantInt(0),
+                                              name("vecx"));
+   res = m_builder->CreateInsertElement(res, x[0], constantInt(1),
+                               name("vecxx"));
+   res = m_builder->CreateInsertElement(res, x[0], constantInt(2),
+                               name("vecxxx"));
+   res = m_builder->CreateInsertElement(res, x[0], constantInt(3),
+                               name("vecxxxx"));
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::constElement(llvm::IRBuilder<>* m_builder, llvm::Value *idx)
+{
+   llvm::Value* res;
+   std::vector<llvm::Value*> res2(4);
    llvm::Value *xChannel, *yChannel, *zChannel, *wChannel;
 
    xChannel = elementPointer(m_consts, idx, 0);
-   yChannel = elementPointer(m_consts, idx, 1);
-   zChannel = elementPointer(m_consts, idx, 2);
-   wChannel = elementPointer(m_consts, idx, 3);
 
-   res[0] = alignedArrayLoad(xChannel);
-   res[1] = alignedArrayLoad(yChannel);
-   res[2] = alignedArrayLoad(zChannel);
-   res[3] = alignedArrayLoad(wChannel);
+   res = alignedArrayLoad(xChannel);
 
-   return res;
+   res2[0]=unpackConstElement(m_builder, res,0);
+   res2[1]=unpackConstElement(m_builder, res,1);
+   res2[2]=unpackConstElement(m_builder, res,2);
+   res2[3]=unpackConstElement(m_builder, res,3);
+
+   return res2;
 }
 
 std::vector<llvm::Value*> StorageSoa::outputElement(llvm::Value *idx)
@@ -260,6 +281,12 @@ llvm::Module * StorageSoa::currentModule() const
     return m_block->getParent()->getParent();
 }
 
+llvm::Constant * StorageSoa::createConstGlobalFloat(const float val)
+{
+   Constant*c = ConstantFP::get(APFloat(val));
+   return c;
+}
+
 llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &vec)
 {
    VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
@@ -278,7 +305,7 @@ llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &v
 }
 
 std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, int swizzle,
-                                           llvm::Value *indIdx)
+                                           llvm::IRBuilder<>* m_builder,llvm::Value *indIdx)
 {
    std::vector<llvm::Value*> val(4);
 
@@ -302,7 +329,7 @@ std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, in
       val = tempElement(realIndex);
       break;
    case TGSI_FILE_CONSTANT:
-      val = constElement(realIndex);
+      val = constElement(m_builder, realIndex);
       break;
    case TGSI_FILE_IMMEDIATE:
       val = immediateElement(realIndex);
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.h b/src/gallium/auxiliary/gallivm/storagesoa.h
index ae2fc7c6ae..f21ca6ec43 100644
--- a/src/gallium/auxiliary/gallivm/storagesoa.h
+++ b/src/gallium/auxiliary/gallivm/storagesoa.h
@@ -29,6 +29,7 @@
 #define STORAGESOA_H
 
 #include <pipe/p_shader_tokens.h>
+#include <llvm/Support/IRBuilder.h>
 
 #include <vector>
 #include <list>
@@ -56,7 +57,7 @@ public:
 
 
    std::vector<llvm::Value*> load(enum tgsi_file_type type, int idx, int swizzle, 
-                                  llvm::Value *indIdx =0);
+                                  llvm::IRBuilder<>* m_builder, llvm::Value *indIdx =0);
    void store(enum tgsi_file_type type, int idx, const std::vector<llvm::Value*> &val,
               int mask);
 
@@ -76,10 +77,12 @@ private:
    const char *name(const char *prefix) const;
    llvm::Value  *alignedArrayLoad(llvm::Value *val);
    llvm::Module *currentModule() const;
+   llvm::Constant  *createConstGlobalFloat(const float val);
    llvm::Constant  *createConstGlobalVector(const std::vector<float> &vec);
 
    std::vector<llvm::Value*> inputElement(llvm::Value *indIdx);
-   std::vector<llvm::Value*> constElement(llvm::Value *indIdx);
+   llvm::Value* unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx, int cc);
+   std::vector<llvm::Value*> constElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx);
    std::vector<llvm::Value*> outputElement(llvm::Value *indIdx);
    std::vector<llvm::Value*> tempElement(llvm::Value *indIdx);
    std::vector<llvm::Value*> immediateElement(llvm::Value *indIdx);
diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
index cc1516a45e..1191a6cae9 100644
--- a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
@@ -52,7 +52,7 @@ static inline FunctionType *vertexShaderFunctionType()
    // pass are castable to the following:
    // [4 x <4 x float>] inputs,
    // [4 x <4 x float>] output,
-   // [4 x [4 x float]] consts,
+   // [4 x [1 x float]] consts,
    // [4 x <4 x float>] temps
 
    std::vector<const Type*> funcArgs;
@@ -61,7 +61,7 @@ static inline FunctionType *vertexShaderFunctionType()
    PointerType *vectorArrayPtr = PointerType::get(vectorArray, 0);
 
    ArrayType   *floatArray     = ArrayType::get(Type::FloatTy, 4);
-   ArrayType   *constsArray    = ArrayType::get(floatArray, 4);
+   ArrayType   *constsArray    = ArrayType::get(floatArray, 1);
    PointerType *constsArrayPtr = PointerType::get(constsArray, 0);
 
    funcArgs.push_back(vectorArrayPtr);//inputs
@@ -246,6 +246,7 @@ translate_instruction(llvm::Module *module,
          val = storage->constElement(src->SrcRegister.Index, indIdx);
       } else if (src->SrcRegister.File == TGSI_FILE_INPUT) {
          val = storage->inputElement(src->SrcRegister.Index, indIdx);
+      // FIXME we should not be generating elements for temporaries, this creates useless memory writes
       } else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) {
          val = storage->tempElement(src->SrcRegister.Index);
       } else if (src->SrcRegister.File == TGSI_FILE_OUTPUT) {
@@ -286,9 +287,13 @@ translate_instruction(llvm::Module *module,
       out = instr->rsq(inputs[0]);
    }
       break;
-   case TGSI_OPCODE_EXP:
+   case TGSI_OPCODE_EXP: {
+      out = instr->exp(inputs[0]);
+   }
       break;
-   case TGSI_OPCODE_LOG:
+   case TGSI_OPCODE_LOG: {
+      out = instr->log(inputs[0]);
+   }
       break;
    case TGSI_OPCODE_MUL: {
       out = instr->mul(inputs[0], inputs[1]);
@@ -338,21 +343,31 @@ translate_instruction(llvm::Module *module,
       out = instr->lerp(inputs[0], inputs[1], inputs[2]);
    }
       break;
-   case TGSI_OPCODE_CND:
+   case TGSI_OPCODE_CND: {
+      out = instr->cnd(inputs[0], inputs[1], inputs[2]);
+   }
       break;
-   case TGSI_OPCODE_CND0:
+   case TGSI_OPCODE_CND0: {
+      out = instr->cnd0(inputs[0], inputs[1], inputs[2]);
+   }
       break;
-   case TGSI_OPCODE_DOT2ADD:
+   case TGSI_OPCODE_DOT2ADD: {
+      out = instr->dot2add(inputs[0], inputs[1], inputs[2]);
+   }
       break;
    case TGSI_OPCODE_INDEX:
       break;
-   case TGSI_OPCODE_NEGATE:
+   case TGSI_OPCODE_NEGATE: {
+      out = instr->neg(inputs[0]);
+   }
       break;
    case TGSI_OPCODE_FRAC: {
       out = instr->frc(inputs[0]);
    }
       break;
-   case TGSI_OPCODE_CLAMP:
+   case TGSI_OPCODE_CLAMP: {
+      out = instr->clamp(inputs[0]);
+   }
       break;
    case TGSI_OPCODE_FLOOR: {
       out = instr->floor(inputs[0]);
@@ -392,9 +407,13 @@ translate_instruction(llvm::Module *module,
       out = instr->cos(inputs[0]);
    }
       break;
-   case TGSI_OPCODE_DDX:
+   case TGSI_OPCODE_DDX: {
+      out = instr->ddx(inputs[0]);
+   }
       break;
-   case TGSI_OPCODE_DDY:
+   case TGSI_OPCODE_DDY: {
+      out = instr->ddy(inputs[0]);
+   }
       break;
    case TGSI_OPCODE_KILP:
       break;
@@ -408,9 +427,13 @@ translate_instruction(llvm::Module *module,
       break;
    case TGSI_OPCODE_RFL:
       break;
-   case TGSI_OPCODE_SEQ:
+   case TGSI_OPCODE_SEQ: {
+      out = instr->seq(inputs[0], inputs[1]);
+   }
       break;
-   case TGSI_OPCODE_SFL:
+   case TGSI_OPCODE_SFL: {
+      out = instr->sfl(inputs[0], inputs[1]);
+   }
       break;
    case TGSI_OPCODE_SGT: {
       out = instr->sgt(inputs[0], inputs[1]);
@@ -420,11 +443,17 @@ translate_instruction(llvm::Module *module,
       out = instr->sin(inputs[0]);
    }
       break;
-   case TGSI_OPCODE_SLE:
+   case TGSI_OPCODE_SLE: {
+      out = instr->sle(inputs[0], inputs[1]);
+   }
       break;
-   case TGSI_OPCODE_SNE:
+   case TGSI_OPCODE_SNE: {
+      out = instr->sne(inputs[0], inputs[1]);
+   }
       break;
-   case TGSI_OPCODE_STR:
+   case TGSI_OPCODE_STR: {
+      out = instr->str(inputs[0], inputs[1]);
+   }
       break;
    case TGSI_OPCODE_TEX:
       break;
@@ -438,7 +467,9 @@ translate_instruction(llvm::Module *module,
       break;
    case TGSI_OPCODE_UP4UB:
       break;
-   case TGSI_OPCODE_X2D:
+   case TGSI_OPCODE_X2D: {
+      out = instr->x2d(inputs[0], inputs[1], inputs[2]);
+   }
       break;
    case TGSI_OPCODE_ARA:
       break;
@@ -468,11 +499,18 @@ translate_instruction(llvm::Module *module,
       break;
    case TGSI_OPCODE_TXB:
       break;
-   case TGSI_OPCODE_NRM:
+   case TGSI_OPCODE_NRM4:
+   case TGSI_OPCODE_NRM: {
+      out = instr->nrm(inputs[0]);
+   }
       break;
-   case TGSI_OPCODE_DIV:
+   case TGSI_OPCODE_DIV: {
+      out = instr->div(inputs[0], inputs[1]);
+   }
       break;
-   case TGSI_OPCODE_DP2:
+   case TGSI_OPCODE_DP2: {
+      out = instr->dp2(inputs[0], inputs[1]);
+   }
       break;
    case TGSI_OPCODE_TXL:
       break;
@@ -590,8 +628,6 @@ translate_instruction(llvm::Module *module,
       break;
    case TGSI_OPCODE_M3X2:
       break;
-   case TGSI_OPCODE_NRM4:
-      break;
    case TGSI_OPCODE_CALLNZ:
       break;
    case TGSI_OPCODE_IFC:
@@ -641,6 +677,7 @@ translate_instruction(llvm::Module *module,
 
       if (dst->DstRegister.File == TGSI_FILE_OUTPUT) {
          storage->setOutputElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      // FIXME we should not be generating elements for temporaries, this creates useless memory writes
       } else if (dst->DstRegister.File == TGSI_FILE_TEMPORARY) {
          storage->setTempElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
       } else if (dst->DstRegister.File == TGSI_FILE_ADDRESS) {
@@ -672,9 +709,8 @@ translate_instructionir(llvm::Module *module,
       if (src->SrcRegister.Indirect) {
          indIdx = storage->addrElement(src->SrcRegisterInd.Index);
       }
-
       val = storage->load((enum tgsi_file_type)src->SrcRegister.File,
-                          src->SrcRegister.Index, swizzle, indIdx);
+                          src->SrcRegister.Index, swizzle, instr->getIRBuilder(), indIdx);
 
       inputs[i] = val;
    }
@@ -732,6 +768,7 @@ translate_instructionir(llvm::Module *module,
    }
       break;
    case TGSI_OPCODE_SLT: {
+      out = instr->slt(inputs[0], inputs[1]);
    }
       break;
    case TGSI_OPCODE_SGE: {
@@ -989,7 +1026,6 @@ translate_instructionir(llvm::Module *module,
    /* store results  */
    for (int i = 0; i < inst->Instruction.NumDstRegs; ++i) {
       struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
-
       storage->store((enum tgsi_file_type)dst->DstRegister.File,
                      dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
    }
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
index 491141f190..dea1aed032 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
@@ -164,6 +164,27 @@ rem_prefix(const char *longname)
 }
 
 
+static const char *
+reg_name(int reg)
+{
+   switch (reg) {
+   case SPE_REG_SP:
+      return "$sp";
+   case SPE_REG_RA:
+      return "$lr";
+   default:
+      {
+         /* cycle through four buffers to handle multiple calls per printf */
+         static char buf[4][10];
+         static int b = 0;
+         b = (b + 1) % 4;
+         sprintf(buf[b], "$%d", reg);
+         return buf[b];
+      }
+   }
+}
+
+
 static void emit_RR(struct spe_function *p, unsigned op, unsigned rT,
 		    unsigned rA, unsigned rB, const char *name)
 {
@@ -176,7 +197,8 @@ static void emit_RR(struct spe_function *p, unsigned op, unsigned rT,
     assert(p->num_inst <= p->max_inst);
     if (p->print) {
        indent(p);
-       printf("%s\t$%d, $%d, $%d\n", rem_prefix(name), rT, rA, rB);
+       printf("%s\t%s, %s, %s\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), reg_name(rB));
     }
 }
 
@@ -194,7 +216,8 @@ static void emit_RRR(struct spe_function *p, unsigned op, unsigned rT,
     assert(p->num_inst <= p->max_inst);
     if (p->print) {
        indent(p);
-       printf("%s\t$%d, $%d, $%d, $%d\n", rem_prefix(name), rT, rA, rB, rC);
+       printf("%s\t%s, %s, %s, %s\n", rem_prefix(name), reg_name(rT),
+              reg_name(rA), reg_name(rB), reg_name(rC));
     }
 }
 
@@ -211,7 +234,8 @@ static void emit_RI7(struct spe_function *p, unsigned op, unsigned rT,
     assert(p->num_inst <= p->max_inst);
     if (p->print) {
        indent(p);
-       printf("%s\t$%d, $%d, 0x%x\n", rem_prefix(name), rT, rA, imm);
+       printf("%s\t%s, %s, 0x%x\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), imm);
     }
 }
 
@@ -229,7 +253,8 @@ static void emit_RI8(struct spe_function *p, unsigned op, unsigned rT,
     assert(p->num_inst <= p->max_inst);
     if (p->print) {
        indent(p);
-       printf("%s\t$%d, $%d, 0x%x\n", rem_prefix(name), rT, rA, imm);
+       printf("%s\t%s, %s, 0x%x\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), imm);
     }
 }
 
@@ -247,15 +272,22 @@ static void emit_RI10(struct spe_function *p, unsigned op, unsigned rT,
     assert(p->num_inst <= p->max_inst);
     if (p->print) {
        indent(p);
-       if (strcmp(name, "spe_lqd") == 0 ||
-           strcmp(name, "spe_stqd") == 0)
-          printf("%s\t$%d, 0x%x($%d)\n", rem_prefix(name), rT, imm, rA);
-       else
-          printf("%s\t$%d, $%d, 0x%x\n", rem_prefix(name), rT, rA, imm);
+       printf("%s\t%s, %s, 0x%x\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), imm);
     }
 }
 
 
+/** As above, but do range checking on signed immediate value */
+static void emit_RI10s(struct spe_function *p, unsigned op, unsigned rT,
+                       unsigned rA, int imm, const char *name)
+{
+    assert(imm <= 511);
+    assert(imm >= -512);
+    emit_RI10(p, op, rT, rA, imm, name);
+}
+
+
 static void emit_RI16(struct spe_function *p, unsigned op, unsigned rT,
 		      int imm, const char *name)
 {
@@ -267,7 +299,7 @@ static void emit_RI16(struct spe_function *p, unsigned op, unsigned rT,
     assert(p->num_inst <= p->max_inst);
     if (p->print) {
        indent(p);
-       printf("%s\t$%d, 0x%x\n", rem_prefix(name), rT, imm);
+       printf("%s\t%s, 0x%x\n", rem_prefix(name), reg_name(rT), imm);
     }
 }
 
@@ -283,7 +315,7 @@ static void emit_RI18(struct spe_function *p, unsigned op, unsigned rT,
     assert(p->num_inst <= p->max_inst);
     if (p->print) {
        indent(p);
-       printf("%s\t$%d, 0x%x\n", rem_prefix(name), rT, imm);
+       printf("%s\t%s, 0x%x\n", rem_prefix(name), reg_name(rT), imm);
     }
 }
 
@@ -332,6 +364,12 @@ void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \
    emit_RI10(p, _op, rT, rA, imm, __FUNCTION__);             \
 }
 
+#define EMIT_RI10s(_name, _op) \
+void _name (struct spe_function *p, unsigned rT, unsigned rA, int imm) \
+{ \
+   emit_RI10s(p, _op, rT, rA, imm, __FUNCTION__);             \
+}
+
 #define EMIT_RI16(_name, _op) \
 void _name (struct spe_function *p, unsigned rT, int imm) \
 { \
@@ -353,20 +391,28 @@ void _name (struct spe_function *p, int imm) \
 #include "rtasm_ppc_spe.h"
 
 
+
 /**
  * Initialize an spe_function.
  * \param code_size  size of instruction buffer to allocate, in bytes.
  */
 void spe_init_func(struct spe_function *p, unsigned code_size)
 {
+    unsigned int i;
+
     p->store = align_malloc(code_size, 16);
     p->num_inst = 0;
     p->max_inst = code_size / SPE_INST_SIZE;
 
+    p->set_count = 0;
+    memset(p->regs, 0, SPE_NUM_REGS * sizeof(p->regs[0]));
+
     /* Conservatively treat R0 - R2 and R80 - R127 as non-volatile.
      */
-    p->regs[0] = ~7;
-    p->regs[1] = (1U << (80 - 64)) - 1;
+    p->regs[0] = p->regs[1] = p->regs[2] = 1;
+    for (i = 80; i <= 127; i++) {
+      p->regs[i] = 1;
+    }
 
     p->print = false;
     p->indent = 0;
@@ -398,12 +444,8 @@ int spe_allocate_available_register(struct spe_function *p)
 {
    unsigned i;
    for (i = 0; i < SPE_NUM_REGS; i++) {
-      const uint64_t mask = (1ULL << (i % 64));
-      const unsigned idx = i / 64;
-
-      assert(idx < 2);
-      if ((p->regs[idx] & mask) != 0) {
-         p->regs[idx] &= ~mask;
+      if (p->regs[i] == 0) {
+         p->regs[i] = 1;
          return i;
       }
    }
@@ -417,31 +459,84 @@ int spe_allocate_available_register(struct spe_function *p)
  */
 int spe_allocate_register(struct spe_function *p, int reg)
 {
-   const unsigned idx = reg / 64;
-   const unsigned bit = reg % 64;
-
    assert(reg < SPE_NUM_REGS);
-   assert((p->regs[idx] & (1ULL << bit)) != 0);
-
-   p->regs[idx] &= ~(1ULL << bit);
+   assert(p->regs[reg] == 0);
+   p->regs[reg] = 1;
    return reg;
 }
 
 
 /**
- * Mark the given SPE register as "unallocated".
+ * Mark the given SPE register as "unallocated".  Note that this should
+ * only be used on registers allocated in the current register set; an
+ * assertion will fail if an attempt is made to deallocate a register
+ * allocated in an earlier register set.
  */
 void spe_release_register(struct spe_function *p, int reg)
 {
-   const unsigned idx = reg / 64;
-   const unsigned bit = reg % 64;
+   assert(reg < SPE_NUM_REGS);
+   assert(p->regs[reg] == 1);
 
-   assert(idx < 2);
+   p->regs[reg] = 0;
+}
 
-   assert(reg < SPE_NUM_REGS);
-   assert((p->regs[idx] & (1ULL << bit)) == 0);
+/**
+ * Start a new set of registers.  This can be called if
+ * it will be difficult later to determine exactly what
+ * registers were actually allocated during a code generation
+ * sequence, and you really just want to deallocate all of them.
+ */
+void spe_allocate_register_set(struct spe_function *p)
+{
+   unsigned int i;
 
-   p->regs[idx] |= (1ULL << bit);
+   /* Keep track of the set count.  If it ever wraps around to 0, 
+    * we're in trouble.
+    */
+   p->set_count++;
+   assert(p->set_count > 0);
+
+   /* Increment the allocation count of all registers currently
+    * allocated.  Then any registers that are allocated in this set
+    * will be the only ones with a count of 1; they'll all be released
+    * when the register set is released.
+    */
+   for (i = 0; i < SPE_NUM_REGS; i++) {
+      if (p->regs[i] > 0)
+         p->regs[i]++;
+   }
+}
+
+void spe_release_register_set(struct spe_function *p)
+{
+   unsigned int i;
+
+   /* If the set count drops below zero, we're in trouble. */
+   assert(p->set_count > 0);
+   p->set_count--;
+
+   /* Drop the allocation level of all registers.  Any allocated
+    * during this register set will drop to 0 and then become
+    * available.
+    */
+   for (i = 0; i < SPE_NUM_REGS; i++) {
+      if (p->regs[i] > 0)
+         p->regs[i]--;
+   }
+}
+
+
+unsigned
+spe_get_registers_used(const struct spe_function *p, ubyte used[])
+{
+   unsigned i, num = 0;
+   /* only count registers in the range available to callers */
+   for (i = 2; i < 80; i++) {
+      if (p->regs[i]) {
+         used[num++] = i;
+      }
+   }
+   return num;
 }
 
 
@@ -459,7 +554,7 @@ spe_indent(struct spe_function *p, int spaces)
 }
 
 
-extern void
+void
 spe_comment(struct spe_function *p, int rel_indent, const char *s)
 {
    if (p->print) {
@@ -472,6 +567,56 @@ spe_comment(struct spe_function *p, int rel_indent, const char *s)
 
 
 /**
+ * Load quad word.
+ * NOTE: offset is in bytes and the least significant 4 bits must be zero!
+ */
+void spe_lqd(struct spe_function *p, unsigned rT, unsigned rA, int offset)
+{
+   const boolean pSave = p->print;
+
+   /* offset must be a multiple of 16 */
+   assert(offset % 16 == 0);
+   /* offset must fit in 10-bit signed int field, after shifting */
+   assert((offset >> 4) <= 511);
+   assert((offset >> 4) >= -512);
+
+   p->print = FALSE;
+   emit_RI10(p, 0x034, rT, rA, offset >> 4, "spe_lqd");
+   p->print = pSave;
+
+   if (p->print) {
+      indent(p);
+      printf("lqd\t%s, %d(%s)\n", reg_name(rT), offset, reg_name(rA));
+   }
+}
+
+
+/**
+ * Store quad word.
+ * NOTE: offset is in bytes and the least significant 4 bits must be zero!
+ */
+void spe_stqd(struct spe_function *p, unsigned rT, unsigned rA, int offset)
+{
+   const boolean pSave = p->print;
+
+   /* offset must be a multiple of 16 */
+   assert(offset % 16 == 0);
+   /* offset must fit in 10-bit signed int field, after shifting */
+   assert((offset >> 4) <= 511);
+   assert((offset >> 4) >= -512);
+
+   p->print = FALSE;
+   emit_RI10(p, 0x024, rT, rA, offset >> 4, "spe_stqd");
+   p->print = pSave;
+
+   if (p->print) {
+      indent(p);
+      printf("stqd\t%s, %d(%s)\n", reg_name(rT), offset, reg_name(rA));
+   }
+}
+
+
+/**
  * For branch instructions:
  * \param d  if 1, disable interupts if branch is taken
  * \param e  if 1, enable interupts if branch is taken
@@ -603,22 +748,187 @@ void spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui)
 {
    /* If the whole value is in the lower 18 bits, use ila, which
     * doesn't sign-extend.  Otherwise, if the two halfwords of
-    * the constant are identical, use ilh.  Otherwise, we have
-    * to use ilhu followed by iohl.
+    * the constant are identical, use ilh.  Otherwise, if every byte of
+    * the desired value is 0x00 or 0xff, we can use Form Select Mask for
+    * Bytes Immediate (fsmbi) to load the value in a single instruction.
+    * Otherwise, in the general case, we have to use ilhu followed by iohl.
     */
-   if ((ui & 0xfffc0000) == ui) {
+   if ((ui & 0x0003ffff) == ui) {
       spe_ila(p, rT, ui);
    }
    else if ((ui >> 16) == (ui & 0xffff)) {
       spe_ilh(p, rT, ui & 0xffff);
    }
+   else if (
+      ((ui & 0x000000ff) == 0 || (ui & 0x000000ff) == 0x000000ff) &&
+      ((ui & 0x0000ff00) == 0 || (ui & 0x0000ff00) == 0x0000ff00) &&
+      ((ui & 0x00ff0000) == 0 || (ui & 0x00ff0000) == 0x00ff0000) &&
+      ((ui & 0xff000000) == 0 || (ui & 0xff000000) == 0xff000000)
+   ) {
+      unsigned int mask = 0;
+      /* fsmbi duplicates each bit in the given mask eight times,
+       * using a 16-bit value to initialize a 16-byte quadword.
+       * Each 4-bit nybble of the mask corresponds to a full word
+       * of the result; look at the value and figure out the mask
+       * (replicated for each word in the quadword), and then
+       * form the "select mask" to get the value.
+       */
+      if ((ui & 0x000000ff) == 0x000000ff) mask |= 0x1111;
+      if ((ui & 0x0000ff00) == 0x0000ff00) mask |= 0x2222;
+      if ((ui & 0x00ff0000) == 0x00ff0000) mask |= 0x4444;
+      if ((ui & 0xff000000) == 0xff000000) mask |= 0x8888;
+      spe_fsmbi(p, rT, mask);
+   }
    else {
+      /* The general case: this usually uses two instructions, but
+       * may use only one if the low-order 16 bits of each word are 0.
+       */
       spe_ilhu(p, rT, ui >> 16);
       if (ui & 0xffff)
          spe_iohl(p, rT, ui & 0xffff);
    }
 }
 
+/**
+ * This function is constructed identically to spe_xor_uint() below.
+ * Changes to one should be made in the other.
+ */
+void
+spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+{
+   /* If we can, emit a single instruction, either And Byte Immediate
+    * (which uses the same constant across each byte), And Halfword Immediate
+    * (which sign-extends a 10-bit immediate to 16 bits and uses that
+    * across each halfword), or And Word Immediate (which sign-extends
+    * a 10-bit immediate to 32 bits).
+    *
+    * Otherwise, we'll need to use a temporary register.
+    */
+   unsigned int tmp;
+
+   /* If the upper 23 bits are all 0s or all 1s, sign extension
+    * will work and we can use And Word Immediate
+    */
+   tmp = ui & 0xfffffe00;
+   if (tmp == 0xfffffe00 || tmp  == 0) {
+      spe_andi(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+   
+   /* If the ui field is symmetric along halfword boundaries and
+    * the upper 7 bits of each halfword are all 0s or 1s, we
+    * can use And Halfword Immediate
+    */
+   tmp = ui & 0xfe00fe00;
+   if ((tmp == 0xfe00fe00 || tmp == 0) && ((ui >> 16) == (ui & 0x0000ffff))) {
+      spe_andhi(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+
+   /* If the ui field is symmetric in each byte, then we can use
+    * the And Byte Immediate instruction.
+    */
+   tmp = ui & 0x000000ff;
+   if ((ui >> 24) == tmp && ((ui >> 16) & 0xff) == tmp && ((ui >> 8) & 0xff) == tmp) {
+      spe_andbi(p, rT, rA, tmp);
+      return;
+   }
+
+   /* Otherwise, we'll have to use a temporary register. */
+   unsigned int tmp_reg = spe_allocate_available_register(p);
+   spe_load_uint(p, tmp_reg, ui);
+   spe_and(p, rT, rA, tmp_reg);
+   spe_release_register(p, tmp_reg);
+}
+
+
+/**
+ * This function is constructed identically to spe_and_uint() above.
+ * Changes to one should be made in the other.
+ */
+void
+spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+{
+   /* If we can, emit a single instruction, either Exclusive Or Byte 
+    * Immediate (which uses the same constant across each byte), Exclusive 
+    * Or Halfword Immediate (which sign-extends a 10-bit immediate to 
+    * 16 bits and uses that across each halfword), or Exclusive Or Word 
+    * Immediate (which sign-extends a 10-bit immediate to 32 bits).
+    *
+    * Otherwise, we'll need to use a temporary register.
+    */
+   unsigned int tmp;
+
+   /* If the upper 23 bits are all 0s or all 1s, sign extension
+    * will work and we can use Exclusive Or Word Immediate
+    */
+   tmp = ui & 0xfffffe00;
+   if (tmp == 0xfffffe00 || tmp  == 0) {
+      spe_xori(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+   
+   /* If the ui field is symmetric along halfword boundaries and
+    * the upper 7 bits of each halfword are all 0s or 1s, we
+    * can use Exclusive Or Halfword Immediate
+    */
+   tmp = ui & 0xfe00fe00;
+   if ((tmp == 0xfe00fe00 || tmp == 0) && ((ui >> 16) == (ui & 0x0000ffff))) {
+      spe_xorhi(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+
+   /* If the ui field is symmetric in each byte, then we can use
+    * the Exclusive Or Byte Immediate instruction.
+    */
+   tmp = ui & 0x000000ff;
+   if ((ui >> 24) == tmp && ((ui >> 16) & 0xff) == tmp && ((ui >> 8) & 0xff) == tmp) {
+      spe_xorbi(p, rT, rA, tmp);
+      return;
+   }
+
+   /* Otherwise, we'll have to use a temporary register. */
+   unsigned int tmp_reg = spe_allocate_available_register(p);
+   spe_load_uint(p, tmp_reg, ui);
+   spe_xor(p, rT, rA, tmp_reg);
+   spe_release_register(p, tmp_reg);
+}
+
+void
+spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+{
+   /* If the comparison value is 9 bits or less, it fits inside a
+    * Compare Equal Word Immediate instruction.
+    */
+   if ((ui & 0x000001ff) == ui) {
+      spe_ceqi(p, rT, rA, ui);
+   }
+   /* Otherwise, we're going to have to load a word first. */
+   else {
+      unsigned int tmp_reg = spe_allocate_available_register(p);
+      spe_load_uint(p, tmp_reg, ui);
+      spe_ceq(p, rT, rA, tmp_reg);
+      spe_release_register(p, tmp_reg);
+   }
+}
+
+void
+spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui)
+{
+   /* If the comparison value is 10 bits or less, it fits inside a
+    * Compare Logical Greater Than Word Immediate instruction.
+    */
+   if ((ui & 0x000003ff) == ui) {
+      spe_clgti(p, rT, rA, ui);
+   }
+   /* Otherwise, we're going to have to load a word first. */
+   else {
+      unsigned int tmp_reg = spe_allocate_available_register(p);
+      spe_load_uint(p, tmp_reg, ui);
+      spe_clgt(p, rT, rA, tmp_reg);
+      spe_release_register(p, tmp_reg);
+   }
+}
 
 void
 spe_splat(struct spe_function *p, unsigned rT, unsigned rA)
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
index 61c7edeb60..d6a3c02f20 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
@@ -53,17 +53,26 @@ struct spe_function
    uint num_inst;
    uint max_inst;
 
-    /**
-     * Mask of used / unused registers
-     *
-     * Each set bit corresponds to an available register.  Each cleared bit
-     * corresponds to an allocated register.
+   /**
+    * The "set count" reflects the number of nested register sets
+    * are allowed.  In the unlikely case that we exceed the set count,
+    * register allocation will start to be confused, which is critical
+    * enough that we check for it.
+    */
+   unsigned char set_count;
+
+   /** 
+    * Flags for used and unused registers.  Each byte corresponds to a
+    * register; a 0 in that byte means that the register is available.
+    * A value of 1 means that the register was allocated in the current
+    * register set.  Any other value N means that the register was allocated
+    * N register sets ago.
      *
      * \sa
      * spe_allocate_register, spe_allocate_available_register,
-     * spe_release_register
+     * spe_allocate_register_set, spe_release_register_set, spe_release_register, 
      */
-    uint64_t regs[SPE_NUM_REGS / 64];
+    unsigned char regs[SPE_NUM_REGS];
 
     boolean print; /**< print/dump instructions as they're emitted? */
     int indent;    /**< number of spaces to indent */
@@ -77,6 +86,11 @@ extern unsigned spe_code_size(const struct spe_function *p);
 extern int spe_allocate_available_register(struct spe_function *p);
 extern int spe_allocate_register(struct spe_function *p, int reg);
 extern void spe_release_register(struct spe_function *p, int reg);
+extern void spe_allocate_register_set(struct spe_function *p);
+extern void spe_release_register_set(struct spe_function *p);
+
+extern unsigned
+spe_get_registers_used(const struct spe_function *p, ubyte used[]);
 
 extern void spe_print_code(struct spe_function *p, boolean enable);
 extern void spe_indent(struct spe_function *p, int spaces);
@@ -105,6 +119,9 @@ extern void spe_comment(struct spe_function *p, int rel_indent, const char *s);
 #define EMIT_RI10(_name, _op) \
     extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
 			   int imm)
+#define EMIT_RI10s(_name, _op) \
+    extern void _name (struct spe_function *p, unsigned rT, unsigned rA, \
+			   int imm)
 #define EMIT_RI16(_name, _op) \
     extern void _name (struct spe_function *p, unsigned rT, int imm)
 #define EMIT_RI18(_name, _op) \
@@ -117,11 +134,9 @@ extern void spe_comment(struct spe_function *p, int rel_indent, const char *s);
 
 /* Memory load / store instructions
  */
-EMIT_RI10(spe_lqd,  0x034);
 EMIT_RR  (spe_lqx,  0x1c4);
 EMIT_RI16(spe_lqa,  0x061);
 EMIT_RI16(spe_lqr,  0x067);
-EMIT_RI10(spe_stqd, 0x024);
 EMIT_RR  (spe_stqx, 0x144);
 EMIT_RI16(spe_stqa, 0x041);
 EMIT_RI16(spe_stqr, 0x047);
@@ -151,7 +166,7 @@ EMIT_RI16(spe_fsmbi, 0x065);
 EMIT_RR  (spe_ah,      0x0c8);
 EMIT_RI10(spe_ahi,     0x01d);
 EMIT_RR  (spe_a,       0x0c0);
-EMIT_RI10(spe_ai,      0x01c);
+EMIT_RI10s(spe_ai,      0x01c);
 EMIT_RR  (spe_sfh,     0x048);
 EMIT_RI10(spe_sfhi,    0x00d);
 EMIT_RR  (spe_sf,      0x040);
@@ -189,19 +204,19 @@ EMIT_R   (spe_xshw,    0x2ae);
 EMIT_R   (spe_xswd,    0x2a6);
 EMIT_RR  (spe_and,     0x0c1);
 EMIT_RR  (spe_andc,    0x2c1);
-EMIT_RI10(spe_andbi,   0x016);
-EMIT_RI10(spe_andhi,   0x015);
-EMIT_RI10(spe_andi,    0x014);
+EMIT_RI10s(spe_andbi,   0x016);
+EMIT_RI10s(spe_andhi,   0x015);
+EMIT_RI10s(spe_andi,    0x014);
 EMIT_RR  (spe_or,      0x041);
 EMIT_RR  (spe_orc,     0x2c9);
-EMIT_RI10(spe_orbi,    0x006);
-EMIT_RI10(spe_orhi,    0x005);
-EMIT_RI10(spe_ori,     0x004);
+EMIT_RI10s(spe_orbi,    0x006);
+EMIT_RI10s(spe_orhi,    0x005);
+EMIT_RI10s(spe_ori,     0x004);
 EMIT_R   (spe_orx,     0x1f0);
 EMIT_RR  (spe_xor,     0x241);
-EMIT_RI10(spe_xorbi,   0x026);
-EMIT_RI10(spe_xorhi,   0x025);
-EMIT_RI10(spe_xori,    0x024);
+EMIT_RI10s(spe_xorbi,   0x026);
+EMIT_RI10s(spe_xorhi,   0x025);
+EMIT_RI10s(spe_xori,    0x024);
 EMIT_RR  (spe_nand,    0x0c9);
 EMIT_RR  (spe_nor,     0x049);
 EMIT_RR  (spe_eqv,     0x249);
@@ -279,6 +294,12 @@ EMIT_RI16(spe_brz,       0x040);
 EMIT_RI16(spe_brhnz,     0x046);
 EMIT_RI16(spe_brhz,      0x044);
 
+extern void
+spe_lqd(struct spe_function *p, unsigned rT, unsigned rA, int offset);
+
+extern void
+spe_stqd(struct spe_function *p, unsigned rT, unsigned rA, int offset);
+
 extern void spe_bi(struct spe_function *p, unsigned rA, int d, int e);
 extern void spe_iret(struct spe_function *p, unsigned rA, int d, int e);
 extern void spe_bisled(struct spe_function *p, unsigned rT, unsigned rA,
@@ -307,6 +328,22 @@ spe_load_int(struct spe_function *p, unsigned rT, int i);
 extern void
 spe_load_uint(struct spe_function *p, unsigned rT, unsigned int ui);
 
+/** And immediate value into rT. */
+extern void
+spe_and_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+
+/** Xor immediate value into rT. */
+extern void
+spe_xor_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+
+/** Compare equal with immediate value. */
+extern void
+spe_compare_equal_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+
+/** Compare greater with immediate value. */
+extern void
+spe_compare_greater_uint(struct spe_function *p, unsigned rT, unsigned rA, unsigned int ui);
+
 /** Replicate word 0 of rA across rT. */
 extern void
 spe_splat(struct spe_function *p, unsigned rT, unsigned rA);
@@ -388,6 +425,7 @@ EMIT_R   (spe_wrch,       0x10d);
 #undef EMIT_RI7
 #undef EMIT_RI8
 #undef EMIT_RI10
+#undef EMIT_RI10s
 #undef EMIT_RI16
 #undef EMIT_RI18
 #undef EMIT_I16
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
index 3bba9dcc07..99ee74cf14 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -371,7 +371,11 @@ void x86_jcc( struct x86_function *p,
    DUMP_I(cc);
    
    if (offset < 0) {
-      assert(p->csr - p->store > -offset);
+      /*assert(p->csr - p->store > -offset);*/
+      if (p->csr - p->store <= -offset) {
+         /* probably out of memory (using the error_overflow buffer) */
+         return;
+      }
    }
 
    if (offset <= 127 && offset >= -128) {
@@ -675,6 +679,44 @@ void x86_and( struct x86_function *p,
  * SSE instructions
  */
 
+void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 0, ptr);
+}
+
+void sse_prefetch0( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 1, ptr);
+}
+
+void sse_prefetch1( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 2, ptr);
+}
+
+void sse_movntps( struct x86_function *p, 
+                  struct x86_reg dst,
+                  struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+
+   assert(dst.mod != mod_REG);
+   assert(src.mod == mod_REG);
+   emit_2ub(p, 0x0f, 0x2b);
+   emit_modrm(p, src, dst);
+}
+
+
+
 
 void sse_movss( struct x86_function *p,
 		struct x86_reg dst,
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
index 510aa1b0de..1b5eaaca85 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
@@ -185,6 +185,13 @@ void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg ar
 void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 
+
+void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr);
+void sse_prefetch0( struct x86_function *p, struct x86_reg ptr);
+void sse_prefetch1( struct x86_function *p, struct x86_reg ptr);
+
+void sse_movntps( struct x86_function *p, struct x86_reg dst, struct x86_reg src);
+
 void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
 void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
index 4fdad3a5c7..f79170b9d6 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -25,6 +25,10 @@
  * 
  **************************************************************************/
 
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_X86) && defined(PIPE_ARCH_SSE)
+
 #include "pipe/p_debug.h"
 #include "pipe/p_shader_tokens.h"
 #include "util/u_math.h"
@@ -36,8 +40,6 @@
 
 #include "rtasm/rtasm_x86sse.h"
 
-#ifdef PIPE_ARCH_X86
-
 /* for 1/sqrt()
  *
  * This costs about 100fps (close to 10%) in gears:
diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile
index d3951e4e7d..b3d1045a8f 100644
--- a/src/gallium/auxiliary/util/Makefile
+++ b/src/gallium/auxiliary/util/Makefile
@@ -10,6 +10,7 @@ C_SOURCES = \
 	u_gen_mipmap.c \
 	u_handle_table.c \
 	u_hash_table.c \
+	u_keymap.c \
 	u_math.c \
 	u_mm.c \
 	u_rect.c \
diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript
index e65c17b1cc..8a04955a16 100644
--- a/src/gallium/auxiliary/util/SConscript
+++ b/src/gallium/auxiliary/util/SConscript
@@ -11,13 +11,14 @@ util = env.ConvenienceLibrary(
 		'u_gen_mipmap.c',
 		'u_handle_table.c',
 		'u_hash_table.c',
+		'u_keymap.c',
 		'u_math.c',
 		'u_mm.c',
 		'u_rect.c',
 		'u_simple_shaders.c',
 		'u_snprintf.c',
-        'u_stream_stdc.c',
-        'u_stream_wd.c',
+		'u_stream_stdc.c',
+		'u_stream_wd.c',
 		'u_tile.c',
 		'u_time.c',
 	])
diff --git a/src/gallium/auxiliary/util/u_keymap.c b/src/gallium/auxiliary/util/u_keymap.c
new file mode 100644
index 0000000000..01b17ddb1b
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_keymap.c
@@ -0,0 +1,309 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * Key lookup/associative container.
+ *
+ * Like Jose's u_hash_table, based on CSO cache code for now.
+ *
+ * Author: Brian Paul
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_error.h"
+
+#include "cso_cache/cso_hash.h"
+
+#include "util/u_memory.h"
+#include "util/u_keymap.h"
+
+
+struct keymap
+{
+   struct cso_hash *cso;   
+   unsigned key_size;
+   unsigned max_entries; /* XXX not obeyed net */
+   unsigned num_entries;
+   keymap_delete_func delete_func;
+};
+
+
+struct keymap_item
+{
+   void *key, *value;
+};
+
+
+/**
+ * This the default key-delete function used when the client doesn't
+ * provide one.
+ */
+static void
+default_delete_func(const struct keymap *map,
+                    const void *key, void *data, void *user)
+{
+   FREE((void*) data);
+}
+
+
+static INLINE struct keymap_item *
+hash_table_item(struct cso_hash_iter iter)
+{
+   return (struct keymap_item *) cso_hash_iter_data(iter);
+}
+
+
+/**
+ * Return 4-byte hash key for a block of bytes.
+ */
+static unsigned
+hash(const void *key, unsigned keySize)
+{
+   unsigned i, hash;
+
+   keySize /= 4; /* convert from bytes to uints */
+
+   hash = 0;
+   for (i = 0; i < keySize; i++) {
+      hash ^= (i + 1) * ((const unsigned *) key)[i];
+   }
+
+   /*hash = hash ^ (hash >> 11) ^ (hash >> 22);*/
+
+   return hash;
+}
+
+
+/**
+ * Create a new map.
+ * \param keySize  size of the keys in bytes
+ * \param maxEntries  max number of entries to allow (~0 = infinity)
+ * \param deleteFunc  optional callback to call when entries
+ *                    are deleted/replaced
+ */
+struct keymap *
+util_new_keymap(unsigned keySize, unsigned maxEntries,
+                 keymap_delete_func deleteFunc)
+{
+   struct keymap *map = MALLOC_STRUCT(keymap);
+   if (!map)
+      return NULL;
+   
+   map->cso = cso_hash_create();
+   if (!map->cso) {
+      FREE(map);
+      return NULL;
+   }
+   
+   map->max_entries = maxEntries;
+   map->num_entries = 0;
+   map->key_size = keySize;
+   map->delete_func = deleteFunc ? deleteFunc : default_delete_func;
+
+   return map;
+}
+
+
+/**
+ * Delete/free a keymap and all entries.  The deleteFunc that was given at
+ * create time will be called for each entry.
+ * \param user  user-provided pointer passed through to the delete callback
+ */
+void
+util_delete_keymap(struct keymap *map, void *user)
+{
+   util_keymap_remove_all(map, user);
+   cso_hash_delete(map->cso);
+   FREE(map);
+}
+
+
+static INLINE struct cso_hash_iter
+hash_table_find_iter(const struct keymap *map, const void *key,
+                     unsigned key_hash)
+{
+   struct cso_hash_iter iter;
+   struct keymap_item *item;
+   
+   iter = cso_hash_find(map->cso, key_hash);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct keymap_item *) cso_hash_iter_data(iter);
+      if (!memcmp(item->key, key, map->key_size))
+         break;
+      iter = cso_hash_iter_next(iter);
+   }
+   
+   return iter;
+}
+
+
+static INLINE struct keymap_item *
+hash_table_find_item(const struct keymap *map, const void *key,
+                     unsigned key_hash)
+{
+   struct cso_hash_iter iter = hash_table_find_iter(map, key, key_hash);
+   if (cso_hash_iter_is_null(iter)) {
+      return NULL;
+   }
+   else {
+      return hash_table_item(iter);
+   }
+}
+
+
+/**
+ * Insert a new key + data pointer into the table.
+ * Note: we create a copy of the key, but not the data!
+ * If the key is already present in the table, replace the existing
+ * entry (calling the delete callback on the previous entry).
+ * If the maximum capacity of the map is reached an old entry
+ * will be deleted (the delete callback will be called).
+ */
+boolean
+util_keymap_insert(struct keymap *map, const void *key,
+                   const void *data, void *user)
+{
+   unsigned key_hash;
+   struct keymap_item *item;
+   struct cso_hash_iter iter;
+
+   assert(map);
+
+   key_hash = hash(key, map->key_size);
+
+   item = hash_table_find_item(map, key, key_hash);
+   if (item) {
+      /* call delete callback for old entry/item */
+      map->delete_func(map, item->key, item->value, user);
+      item->value = (void *) data;
+      return TRUE;
+   }
+   
+   item = MALLOC_STRUCT(keymap_item);
+   if (!item)
+      return FALSE;
+
+   item->key = mem_dup(key, map->key_size);
+   item->value = (void *) data;
+   
+   iter = cso_hash_insert(map->cso, key_hash, item);
+   if (cso_hash_iter_is_null(iter)) {
+      FREE(item);
+      return FALSE;
+   }
+
+   map->num_entries++;
+
+   return TRUE;
+}
+
+
+/**
+ * Look up a key in the map and return the associated data pointer.
+ */
+const void *
+util_keymap_lookup(const struct keymap *map, const void *key)
+{
+   unsigned key_hash;
+   struct keymap_item *item;
+
+   assert(map);
+
+   key_hash = hash(key, map->key_size);
+
+   item = hash_table_find_item(map, key, key_hash);
+   if (!item)
+      return NULL;
+   
+   return item->value;
+}
+
+
+/**
+ * Remove an entry from the map.
+ * The delete callback will be called if the given key/entry is found.
+ * \param user  passed to the delete callback as the last param.
+ */
+void
+util_keymap_remove(struct keymap *map, const void *key, void *user)
+{
+   unsigned key_hash;
+   struct cso_hash_iter iter;
+   struct keymap_item *item;
+
+   assert(map);
+
+   key_hash = hash(key, map->key_size);
+
+   iter = hash_table_find_iter(map, key, key_hash);
+   if (cso_hash_iter_is_null(iter))
+      return;
+   
+   item = hash_table_item(iter);
+   assert(item);
+   map->delete_func(map, item->key, item->value, user);
+   FREE(item->key);
+   FREE(item);
+   
+   map->num_entries--;
+
+   cso_hash_erase(map->cso, iter);
+}
+
+
+/**
+ * Remove all entries from the map, calling the delete callback for each.
+ * \param user  passed to the delete callback as the last param.
+ */
+void
+util_keymap_remove_all(struct keymap *map, void *user)
+{
+   struct cso_hash_iter iter;
+   struct keymap_item *item;
+
+   assert(map);
+   
+   iter = cso_hash_first_node(map->cso);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct keymap_item *)
+         cso_hash_take(map->cso, cso_hash_iter_key(iter));
+      map->delete_func(map, item->key, item->value, user);
+      FREE(item->key);
+      FREE(item);
+      iter = cso_hash_first_node(map->cso);
+   }
+}
+
+
+extern void
+util_keymap_info(const struct keymap *map)
+{
+   debug_printf("Keymap %p: %u of max %u entries\n",
+                (void *) map, map->num_entries, map->max_entries);
+}
diff --git a/src/gallium/auxiliary/util/u_keymap.h b/src/gallium/auxiliary/util/u_keymap.h
new file mode 100644
index 0000000000..8d60a76fc3
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_keymap.h
@@ -0,0 +1,68 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef U_KEYMAP_H
+#define U_KEYMAP_H
+
+#include "pipe/p_compiler.h"
+
+
+/** opaque keymap type */
+struct keymap;
+
+
+/** Delete/callback function type */
+typedef void (*keymap_delete_func)(const struct keymap *map,
+                                   const void *key, void *data,
+                                   void *user);
+
+
+extern struct keymap *
+util_new_keymap(unsigned keySize, unsigned maxEntries,
+                keymap_delete_func deleteFunc);
+
+extern void
+util_delete_keymap(struct keymap *map, void *user);
+
+extern boolean
+util_keymap_insert(struct keymap *map, const void *key,
+                   const void *data, void *user);
+
+extern const void *
+util_keymap_lookup(const struct keymap *map, const void *key);
+
+extern void
+util_keymap_remove(struct keymap *map, const void *key, void *user);
+
+extern void
+util_keymap_remove_all(struct keymap *map, void *user);
+
+extern void
+util_keymap_info(const struct keymap *map);
+
+
+#endif /* U_KEYMAP_H */
diff --git a/src/gallium/auxiliary/util/u_math.c b/src/gallium/auxiliary/util/u_math.c
index 0729114d6a..5b3cab4642 100644
--- a/src/gallium/auxiliary/util/u_math.c
+++ b/src/gallium/auxiliary/util/u_math.c
@@ -30,7 +30,7 @@
 #include "util/u_math.h"
 
 
-
+/** 2^x, for x in [-1.0, 1.0[ */
 float pow2_table[POW2_TABLE_SIZE];
 
 
@@ -38,9 +38,21 @@ static void
 init_pow2_table(void)
 {
    int i;
-   for (i = 0; i < POW2_TABLE_SIZE; i++) {
-      pow2_table[i] = (float) pow(2.0, i / POW2_TABLE_SCALE);
-   }
+   for (i = 0; i < POW2_TABLE_SIZE; i++)
+      pow2_table[i] = (float) pow(2.0, (i - POW2_TABLE_OFFSET) / POW2_TABLE_SCALE);
+}
+
+
+/** log2(x), for x in [1.0, 2.0[ */
+float log2_table[LOG2_TABLE_SIZE];
+
+
+static void 
+init_log2_table(void)
+{
+   unsigned i;
+   for (i = 0; i < LOG2_TABLE_SIZE; i++)
+      log2_table[i] = (float) log2(1.0 + i * (1.0 / LOG2_TABLE_SIZE));
 }
 
 
@@ -53,6 +65,7 @@ util_init_math(void)
    static boolean initialized = FALSE;
    if (!initialized) {
       init_pow2_table();
+      init_log2_table();
       initialized = TRUE;
    }
 }
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index 084655e6dd..be7303e550 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -174,8 +174,10 @@ static INLINE float logf( float f )
 
 
 
-#define POW2_TABLE_SIZE 256
-#define POW2_TABLE_SCALE ((float) (POW2_TABLE_SIZE-1))
+#define POW2_TABLE_SIZE_LOG2 9
+#define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)
+#define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)
+#define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))
 extern float pow2_table[POW2_TABLE_SIZE];
 
 
@@ -186,98 +188,78 @@ util_init_math(void);
 
 union fi {
    float f;
-   int i;
-   unsigned ui;
+   int32_t i;
+   uint32_t ui;
 };
 
 
 /**
- * Fast approximation to exp(x).
- * Compute with base 2 exponents:  exp(x) = exp2(log2(e) * x)
- * Note: log2(e) is a constant, k = 1.44269
- * So, exp(x) = exp2(k * x);
+ * Fast version of 2^x
  * Identity: exp2(a + b) = exp2(a) * exp2(b)
- * Let ipart = int(k*x)
- * Let fpart = k*x - ipart;
- * So, exp2(k*x) = exp2(ipart) * exp2(fpart)
+ * Let ipart = int(x)
+ * Let fpart = x - ipart;
+ * So, exp2(x) = exp2(ipart) * exp2(fpart)
  * Compute exp2(ipart) with i << ipart
  * Compute exp2(fpart) with lookup table.
  */
 static INLINE float
-util_fast_exp(float x)
+util_fast_exp2(float x)
 {
-   if (x >= 0.0f) {
-      float k = 1.44269f; /* = log2(e) */
-      float kx = k * x;
-      int ipart = (int) kx;
-      float fpart = kx - (float) ipart;
-      float y = (float) (1 << ipart)
-         * pow2_table[(int) (fpart * POW2_TABLE_SCALE)];
-      return y;
-   }
-   else {
-      /* exp(-x) = 1.0 / exp(x) */
-      float k = -1.44269f;
-      float kx = k * x;
-      int ipart = (int) kx;
-      float fpart = kx - (float) ipart;
-      float y = (float) (1 << ipart)
-         * pow2_table[(int) (fpart * POW2_TABLE_SCALE)];
-      return 1.0f / y;
-   }
+   int32_t ipart;
+   float fpart, mpart;
+   union fi epart;
+   
+   if(x > 129.00000f)
+      return 3.402823466e+38f;
+   
+   if(x < -126.99999f)
+      return 0.0f;
+
+   ipart = (int32_t) x;
+   fpart = x - (float) ipart;
+   
+   /* same as
+    *   epart.f = (float) (1 << ipart)
+    * but faster and without integer overflow for ipart > 31 */
+   epart.i = (ipart + 127 ) << 23;
+   
+   mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)];
+   
+   return epart.f * mpart;
 }
 
 
 /**
- * Fast version of 2^x
- * XXX the above function could be implemented in terms of this one.
+ * Fast approximation to exp(x).
  */
 static INLINE float
-util_fast_exp2(float x)
+util_fast_exp(float x)
 {
-   if (x >= 0.0f) {
-      int ipart = (int) x;
-      float fpart = x - (float) ipart;
-      float y = (float) (1 << ipart)
-         * pow2_table[(int) (fpart * POW2_TABLE_SCALE)];
-      return y;
-   }
-   else {
-      /* exp(-x) = 1.0 / exp(x) */
-      int ipart = (int) -x;
-      float fpart = -x - (float) ipart;
-      float y = (float) (1 << ipart)
-         * pow2_table[(int) (fpart * POW2_TABLE_SCALE)];
-      return 1.0f / y;
-   }
+   const float k = 1.44269f; /* = log2(e) */
+   return util_fast_exp2(k * x);
 }
 
 
-/**
- * Based on code from http://www.flipcode.com/archives/Fast_log_Function.shtml
- */
+#define LOG2_TABLE_SIZE_LOG2 8
+#define LOG2_TABLE_SIZE (1 << LOG2_TABLE_SIZE_LOG2)
+extern float log2_table[LOG2_TABLE_SIZE];
+
+
 static INLINE float
-util_fast_log2(float val)
+util_fast_log2(float x)
 {
    union fi num;
-   int log_2;
-   num.f = val;
-   log_2 = ((num.i >> 23) & 255) - 128;
-   num.i &= ~(255 << 23);
-   num.i += 127 << 23;
-   num.f = ((-1.0f/3) * num.f + 2) * num.f - 2.0f/3;
-   return num.f + log_2;
+   float epart, mpart;
+   num.f = x;
+   epart = (float)(((num.i & 0x7f800000) >> 23) - 127);
+   mpart = log2_table[(num.i & 0x007fffff) >> (23 - LOG2_TABLE_SIZE_LOG2)];
+   return epart + mpart;
 }
 
 
 static INLINE float
 util_fast_pow(float x, float y)
 {
-   /* XXX these tests may need adjustment */
-   if (y >= 3.0f && (-0.02f <= x && x <= 0.02f))
-      return 0.0f;
-   if (y >= 50.0f && (-0.9f <= x && x <= 0.9f))
-      return 0.0f;
    return util_fast_exp2(util_fast_log2(x) * y);
 }
 
diff --git a/src/gallium/auxiliary/util/u_sse.h b/src/gallium/auxiliary/util/u_sse.h
index 0c8356cd05..e2a8491e62 100644
--- a/src/gallium/auxiliary/util/u_sse.h
+++ b/src/gallium/auxiliary/util/u_sse.h
@@ -37,6 +37,10 @@
 #ifndef U_SSE_H_
 #define U_SSE_H_
 
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_SSE)
+
 #include <xmmintrin.h>
 #include <emmintrin.h>
 
@@ -66,7 +70,8 @@ _mm_castps_si128(__m128 a)
    return u.m128i;
 }
 
-#endif
+#endif /* defined(_MSC_VER) && _MSC_VER < 1500 */
 
+#endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
 
 #endif /* U_SSE_H_ */