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

Conflicts:
	configs/default

36 files changed, 705 insertions, 488 deletions

diff --git a/src/gallium/drivers/softpipe/sp_clear.c b/src/gallium/drivers/softpipe/sp_clear.c
index 1236706891..dfa46c9fb7 100644
--- a/src/gallium/drivers/softpipe/sp_clear.c
+++ b/src/gallium/drivers/softpipe/sp_clear.c
@@ -31,6 +31,7 @@
 
 
 #include "pipe/p_defines.h"
+#include "util/u_pack_color.h"
 #include "sp_clear.h"
 #include "sp_context.h"
 #include "sp_surface.h"
@@ -39,8 +40,28 @@
 
 
 /**
+ * Convert packed pixel from one format to another.
+ */
+static unsigned
+convert_color(enum pipe_format srcFormat, unsigned srcColor,
+              enum pipe_format dstFormat)
+{
+   ubyte r, g, b, a;
+   unsigned dstColor;
+
+   util_unpack_color_ub(srcFormat, &srcColor, &r, &g, &b, &a);
+   util_pack_color_ub(r, g, b, a, dstFormat, &dstColor);
+
+   return dstColor;
+}
+
+
+
+/**
  * Clear the given surface to the specified value.
  * No masking, no scissor (clear entire buffer).
+ * Note: when clearing a color buffer, the clearValue is always
+ * encoded as PIPE_FORMAT_A8R8G8B8_UNORM.
  */
 void
 softpipe_clear(struct pipe_context *pipe, struct pipe_surface *ps,
@@ -66,7 +87,15 @@ softpipe_clear(struct pipe_context *pipe, struct pipe_surface *ps,
 
    for (i = 0; i < softpipe->framebuffer.num_cbufs; i++) {
       if (ps == sp_tile_cache_get_surface(softpipe->cbuf_cache[i])) {
-         sp_tile_cache_clear(softpipe->cbuf_cache[i], clearValue);
+         unsigned cv;
+         if (ps->format != PIPE_FORMAT_A8R8G8B8_UNORM) {
+            cv = convert_color(PIPE_FORMAT_A8R8G8B8_UNORM, clearValue,
+                               ps->format);
+         }
+         else {
+            cv = clearValue;
+         }
+         sp_tile_cache_clear(softpipe->cbuf_cache[i], cv);
          softpipe->framebuffer.cbufs[i]->status = PIPE_SURFACE_STATUS_CLEAR;
       }
    }
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index 626c3a9d4e..cd1e6663d8 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -32,7 +32,8 @@
 #include "draw/draw_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
-#include "pipe/p_util.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 #include "sp_clear.h"
 #include "sp_context.h"
 #include "sp_flush.h"
@@ -91,17 +92,19 @@ static void softpipe_destroy( struct pipe_context *pipe )
    if (softpipe->draw)
       draw_destroy( softpipe->draw );
 
-   softpipe->quad.polygon_stipple->destroy( softpipe->quad.polygon_stipple );
-   softpipe->quad.earlyz->destroy( softpipe->quad.earlyz );
-   softpipe->quad.shade->destroy( softpipe->quad.shade );
-   softpipe->quad.alpha_test->destroy( softpipe->quad.alpha_test );
-   softpipe->quad.depth_test->destroy( softpipe->quad.depth_test );
-   softpipe->quad.stencil_test->destroy( softpipe->quad.stencil_test );
-   softpipe->quad.occlusion->destroy( softpipe->quad.occlusion );
-   softpipe->quad.coverage->destroy( softpipe->quad.coverage );
-   softpipe->quad.blend->destroy( softpipe->quad.blend );
-   softpipe->quad.colormask->destroy( softpipe->quad.colormask );
-   softpipe->quad.output->destroy( softpipe->quad.output );
+   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
+      softpipe->quad[i].polygon_stipple->destroy( softpipe->quad[i].polygon_stipple );
+      softpipe->quad[i].earlyz->destroy( softpipe->quad[i].earlyz );
+      softpipe->quad[i].shade->destroy( softpipe->quad[i].shade );
+      softpipe->quad[i].alpha_test->destroy( softpipe->quad[i].alpha_test );
+      softpipe->quad[i].depth_test->destroy( softpipe->quad[i].depth_test );
+      softpipe->quad[i].stencil_test->destroy( softpipe->quad[i].stencil_test );
+      softpipe->quad[i].occlusion->destroy( softpipe->quad[i].occlusion );
+      softpipe->quad[i].coverage->destroy( softpipe->quad[i].coverage );
+      softpipe->quad[i].blend->destroy( softpipe->quad[i].blend );
+      softpipe->quad[i].colormask->destroy( softpipe->quad[i].colormask );
+      softpipe->quad[i].output->destroy( softpipe->quad[i].output );
+   }
 
    for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++)
       sp_destroy_tile_cache(softpipe->cbuf_cache[i]);
@@ -112,7 +115,7 @@ static void softpipe_destroy( struct pipe_context *pipe )
 
    for (i = 0; i < Elements(softpipe->constants); i++) {
       if (softpipe->constants[i].buffer) {
-         pipe_buffer_reference(ws, &softpipe->constants[i].buffer, NULL);
+         winsys_buffer_reference(ws, &softpipe->constants[i].buffer, NULL);
       }
    }
 
@@ -128,6 +131,8 @@ softpipe_create( struct pipe_screen *screen,
    struct softpipe_context *softpipe = CALLOC_STRUCT(softpipe_context);
    uint i;
 
+   util_init_math();
+
 #ifdef PIPE_ARCH_X86
    softpipe->use_sse = !debug_get_bool_option( "GALLIUM_NOSSE", FALSE );
 #else
@@ -202,17 +207,19 @@ softpipe_create( struct pipe_screen *screen,
 
 
    /* setup quad rendering stages */
-   softpipe->quad.polygon_stipple = sp_quad_polygon_stipple_stage(softpipe);
-   softpipe->quad.earlyz = sp_quad_earlyz_stage(softpipe);
-   softpipe->quad.shade = sp_quad_shade_stage(softpipe);
-   softpipe->quad.alpha_test = sp_quad_alpha_test_stage(softpipe);
-   softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe);
-   softpipe->quad.stencil_test = sp_quad_stencil_test_stage(softpipe);
-   softpipe->quad.occlusion = sp_quad_occlusion_stage(softpipe);
-   softpipe->quad.coverage = sp_quad_coverage_stage(softpipe);
-   softpipe->quad.blend = sp_quad_blend_stage(softpipe);
-   softpipe->quad.colormask = sp_quad_colormask_stage(softpipe);
-   softpipe->quad.output = sp_quad_output_stage(softpipe);
+   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
+      softpipe->quad[i].polygon_stipple = sp_quad_polygon_stipple_stage(softpipe);
+      softpipe->quad[i].earlyz = sp_quad_earlyz_stage(softpipe);
+      softpipe->quad[i].shade = sp_quad_shade_stage(softpipe);
+      softpipe->quad[i].alpha_test = sp_quad_alpha_test_stage(softpipe);
+      softpipe->quad[i].depth_test = sp_quad_depth_test_stage(softpipe);
+      softpipe->quad[i].stencil_test = sp_quad_stencil_test_stage(softpipe);
+      softpipe->quad[i].occlusion = sp_quad_occlusion_stage(softpipe);
+      softpipe->quad[i].coverage = sp_quad_coverage_stage(softpipe);
+      softpipe->quad[i].blend = sp_quad_blend_stage(softpipe);
+      softpipe->quad[i].colormask = sp_quad_colormask_stage(softpipe);
+      softpipe->quad[i].output = sp_quad_output_stage(softpipe);
+   }
 
    /*
     * Create drawing context and plug our rendering stage into it.
@@ -254,3 +261,4 @@ softpipe_create( struct pipe_screen *screen,
    softpipe_destroy(&softpipe->pipe);
    return NULL;
 }
+
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index 078886f93c..2b9a2a8ee5 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -45,6 +45,10 @@
  */
 #define USE_DRAW_STAGE_PSTIPPLE 1
 
+/* Number of threads working on individual quads.
+ * Setting to 1 disables this feature.
+ */
+#define SP_NUM_QUAD_THREADS 1
 
 struct softpipe_winsys;
 struct softpipe_vbuf_render;
@@ -133,7 +137,7 @@ struct softpipe_context {
       struct quad_stage *output;
 
       struct quad_stage *first; /**< points to one of the above stages */
-   } quad;
+   } quad[SP_NUM_QUAD_THREADS];
 
    /** The primitive drawing context */
    struct draw_context *draw;
@@ -151,13 +155,11 @@ struct softpipe_context {
 };
 
 
-
-
 static INLINE struct softpipe_context *
 softpipe_context( struct pipe_context *pipe )
 {
    return (struct softpipe_context *)pipe;
 }
 
-
 #endif /* SP_CONTEXT_H */
+
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index 12b44a8211..424bd56846 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -34,6 +34,7 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
 #include "pipe/p_winsys.h"
+#include "pipe/p_inlines.h"
 
 #include "sp_context.h"
 #include "sp_state.h"
@@ -135,7 +136,7 @@ softpipe_draw_range_elements(struct pipe_context *pipe,
     */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
       void *buf
-         = pipe->winsys->buffer_map(pipe->winsys,
+         = pipe_buffer_map(pipe->screen,
                                     sp->vertex_buffer[i].buffer,
                                     PIPE_BUFFER_USAGE_CPU_READ);
       draw_set_mapped_vertex_buffer(draw, i, buf);
@@ -143,7 +144,7 @@ softpipe_draw_range_elements(struct pipe_context *pipe,
    /* Map index buffer, if present */
    if (indexBuffer) {
       void *mapped_indexes
-         = pipe->winsys->buffer_map(pipe->winsys, indexBuffer,
+         = pipe_buffer_map(pipe->screen, indexBuffer,
                                     PIPE_BUFFER_USAGE_CPU_READ);
       draw_set_mapped_element_buffer_range(draw, indexSize,
                                            min_index,
@@ -164,11 +165,11 @@ softpipe_draw_range_elements(struct pipe_context *pipe,
     */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
       draw_set_mapped_vertex_buffer(draw, i, NULL);
-      pipe->winsys->buffer_unmap(pipe->winsys, sp->vertex_buffer[i].buffer);
+      pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer);
    }
    if (indexBuffer) {
       draw_set_mapped_element_buffer(draw, 0, NULL);
-      pipe->winsys->buffer_unmap(pipe->winsys, indexBuffer);
+      pipe_buffer_unmap(pipe->screen, indexBuffer);
    }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_flush.c b/src/gallium/drivers/softpipe/sp_flush.c
index e03994b63b..401764bb43 100644
--- a/src/gallium/drivers/softpipe/sp_flush.c
+++ b/src/gallium/drivers/softpipe/sp_flush.c
@@ -73,6 +73,19 @@ softpipe_flush( struct pipe_context *pipe,
       softpipe_unmap_surfaces(softpipe);
    }
 
+   /* Enable to dump BMPs of the color/depth buffers each frame */
+#if 0
+   if(flags & PIPE_FLUSH_FRAME) {
+      static unsigned frame_no = 1;
+      static char filename[256];
+      util_snprintf(filename, sizeof(filename), "cbuf_%u.bmp", frame_no);
+      debug_dump_surface_bmp(filename, softpipe->framebuffer.cbufs[0]);
+      util_snprintf(filename, sizeof(filename), "zsbuf_%u.bmp", frame_no);
+      debug_dump_surface_bmp(filename, softpipe->framebuffer.zsbuf);
+      ++frame_no;
+   }
+#endif
+   
    if (fence)
       *fence = NULL;
 }
diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c
index cc171bbc39..701ee4c72f 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -34,7 +34,7 @@
 
 #include "pipe/p_state.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_inlines.h"
 #include "tgsi/tgsi_exec.h"
 #include "tgsi/tgsi_parse.h"
@@ -106,7 +106,7 @@ exec_run( const struct sp_fragment_shader *base,
 
    /* Compute X, Y, Z, W vals for this quad */
    sp_setup_pos_vector(quad->posCoef, 
-		       (float)quad->x0, (float)quad->y0, 
+		       (float)quad->input.x0, (float)quad->input.y0, 
 		       &machine->QuadPos);
    
    return tgsi_exec_machine_run( machine );
diff --git a/src/gallium/drivers/softpipe/sp_fs_llvm.c b/src/gallium/drivers/softpipe/sp_fs_llvm.c
index 20226da78c..34adac5226 100644
--- a/src/gallium/drivers/softpipe/sp_fs_llvm.c
+++ b/src/gallium/drivers/softpipe/sp_fs_llvm.c
@@ -36,7 +36,7 @@
 
 #include "pipe/p_state.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_inlines.h"
 #include "tgsi/tgsi_sse2.h"
 
diff --git a/src/gallium/drivers/softpipe/sp_fs_sse.c b/src/gallium/drivers/softpipe/sp_fs_sse.c
index 8b7da7c747..496ed43df2 100644
--- a/src/gallium/drivers/softpipe/sp_fs_sse.c
+++ b/src/gallium/drivers/softpipe/sp_fs_sse.c
@@ -34,7 +34,7 @@
 
 #include "pipe/p_state.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_inlines.h"
 #include "tgsi/tgsi_exec.h"
 #include "tgsi/tgsi_sse2.h"
@@ -88,7 +88,7 @@ fs_sse_run( const struct sp_fragment_shader *base,
 
    /* Compute X, Y, Z, W vals for this quad -- place in temp[0] for now */
    sp_setup_pos_vector(quad->posCoef, 
-		       (float)quad->x0, (float)quad->y0, 
+		       (float)quad->input.x0, (float)quad->input.y0, 
 		       machine->Temps);
 
    /* init kill mask */
diff --git a/src/gallium/drivers/softpipe/sp_headers.h b/src/gallium/drivers/softpipe/sp_headers.h
index ae2ee210fc..4a42cb3c19 100644
--- a/src/gallium/drivers/softpipe/sp_headers.h
+++ b/src/gallium/drivers/softpipe/sp_headers.h
@@ -59,20 +59,31 @@
  * Encodes everything we need to know about a 2x2 pixel block.  Uses
  * "Channel-Serial" or "SoA" layout.  
  */
-struct quad_header {
+struct quad_header_input
+{
    int x0;
    int y0;
-   unsigned mask:4;
+   float coverage[QUAD_SIZE];    /** fragment coverage for antialiasing */
    unsigned facing:1;   /**< Front (0) or back (1) facing? */
    unsigned prim:2;     /**< PRIM_POINT, LINE, TRI */
+};
+
+struct quad_header_inout
+{
+   unsigned mask:4;
+};
 
-   struct {
-      /** colors in SOA format (rrrr, gggg, bbbb, aaaa) */
-      float color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE];
-      float depth[QUAD_SIZE];
-   } outputs;
+struct quad_header_output
+{
+   /** colors in SOA format (rrrr, gggg, bbbb, aaaa) */
+   float color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE];
+   float depth[QUAD_SIZE];
+};
 
-   float coverage[QUAD_SIZE];    /** fragment coverage for antialiasing */
+struct quad_header {
+   struct quad_header_input input;
+   struct quad_header_inout inout;
+   struct quad_header_output output;
 
    const struct tgsi_interp_coef *coef;
    const struct tgsi_interp_coef *posCoef;
@@ -80,5 +91,5 @@ struct quad_header {
    unsigned nr_attrs;
 };
 
-
 #endif /* SP_HEADERS_H */
+
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c
index 941ab62e00..038ff04d4f 100644
--- a/src/gallium/drivers/softpipe/sp_prim_setup.c
+++ b/src/gallium/drivers/softpipe/sp_prim_setup.c
@@ -41,7 +41,7 @@
 #include "sp_prim_setup.h"
 #include "draw/draw_pipe.h"
 #include "draw/draw_vertex.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 
 /**
  * Triangle setup info (derived from draw_stage).
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index e9fae951e0..425e13cd28 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -43,6 +43,7 @@
 #include "sp_setup.h"
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
+#include "util/u_memory.h"
 
 
 #define SP_MAX_VBUF_INDEXES 1024
diff --git a/src/gallium/drivers/softpipe/sp_quad.c b/src/gallium/drivers/softpipe/sp_quad.c
index bc83d78ea1..892ef87ee9 100644
--- a/src/gallium/drivers/softpipe/sp_quad.c
+++ b/src/gallium/drivers/softpipe/sp_quad.c
@@ -33,29 +33,33 @@
 static void
 sp_push_quad_first(
    struct softpipe_context *sp,
-   struct quad_stage       *quad )
+   struct quad_stage *quad,
+   uint i )
 {
-   quad->next = sp->quad.first;
-   sp->quad.first = quad;
+   quad->next = sp->quad[i].first;
+   sp->quad[i].first = quad;
 }
 
 static void
 sp_build_depth_stencil(
-   struct softpipe_context *sp )
+   struct softpipe_context *sp,
+   uint i )
 {
    if (sp->depth_stencil->stencil[0].enabled ||
        sp->depth_stencil->stencil[1].enabled) {
-      sp_push_quad_first( sp, sp->quad.stencil_test );
+      sp_push_quad_first( sp, sp->quad[i].stencil_test, i );
    }
    else if (sp->depth_stencil->depth.enabled &&
             sp->framebuffer.zsbuf) {
-      sp_push_quad_first( sp, sp->quad.depth_test );
+      sp_push_quad_first( sp, sp->quad[i].depth_test, i );
    }
 }
 
 void
 sp_build_quad_pipeline(struct softpipe_context *sp)
 {
+   uint i;
+
    boolean early_depth_test =
                sp->depth_stencil->depth.enabled &&
                sp->framebuffer.zsbuf &&
@@ -64,49 +68,51 @@ sp_build_quad_pipeline(struct softpipe_context *sp)
                !sp->fs->info.writes_z;
 
    /* build up the pipeline in reverse order... */
-
-   sp->quad.first = sp->quad.output;
-
-   if (sp->blend->colormask != 0xf) {
-      sp_push_quad_first( sp, sp->quad.colormask );
-   }
-
-   if (sp->blend->blend_enable ||
-       sp->blend->logicop_enable) {
-      sp_push_quad_first( sp, sp->quad.blend );
-   }
-
-   if (sp->depth_stencil->depth.occlusion_count) {
-      sp_push_quad_first( sp, sp->quad.occlusion );
-   }
-
-   if (sp->rasterizer->poly_smooth ||
-       sp->rasterizer->line_smooth ||
-       sp->rasterizer->point_smooth) {
-      sp_push_quad_first( sp, sp->quad.coverage );
-   }
-
-   if (!early_depth_test) {
-      sp_build_depth_stencil( sp );
-   }
-
-   if (sp->depth_stencil->alpha.enabled) {
-      sp_push_quad_first( sp, sp->quad.alpha_test );
-   }
-
-   /* XXX always enable shader? */
-   if (1) {
-      sp_push_quad_first( sp, sp->quad.shade );
-   }
-
-   if (early_depth_test) {
-      sp_build_depth_stencil( sp );
-      sp_push_quad_first( sp, sp->quad.earlyz );
-   }
+   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
+      sp->quad[i].first = sp->quad[i].output;
+
+      if (sp->blend->colormask != 0xf) {
+         sp_push_quad_first( sp, sp->quad[i].colormask, i );
+      }
+
+      if (sp->blend->blend_enable ||
+          sp->blend->logicop_enable) {
+         sp_push_quad_first( sp, sp->quad[i].blend, i );
+      }
+
+      if (sp->depth_stencil->depth.occlusion_count) {
+         sp_push_quad_first( sp, sp->quad[i].occlusion, i );
+      }
+
+      if (sp->rasterizer->poly_smooth ||
+          sp->rasterizer->line_smooth ||
+          sp->rasterizer->point_smooth) {
+         sp_push_quad_first( sp, sp->quad[i].coverage, i );
+      }
+
+      if (!early_depth_test) {
+         sp_build_depth_stencil( sp, i );
+      }
+
+      if (sp->depth_stencil->alpha.enabled) {
+         sp_push_quad_first( sp, sp->quad[i].alpha_test, i );
+      }
+
+      /* XXX always enable shader? */
+      if (1) {
+         sp_push_quad_first( sp, sp->quad[i].shade, i );
+      }
+
+      if (early_depth_test) {
+         sp_build_depth_stencil( sp, i );
+         sp_push_quad_first( sp, sp->quad[i].earlyz, i );
+      }
 
 #if !USE_DRAW_STAGE_PSTIPPLE
-   if (sp->rasterizer->poly_stipple_enable) {
-      sp_push_quad_first( sp, sp->quad.polygon_stipple );
-   }
+      if (sp->rasterizer->poly_stipple_enable) {
+         sp_push_quad_first( sp, sp->quad[i].polygon_stipple, i );
+      }
 #endif
+   }
 }
+
diff --git a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
index 7a42b08ef5..5bebd141e9 100644
--- a/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_alpha_test.c
@@ -7,7 +7,7 @@
 #include "sp_headers.h"
 #include "sp_quad.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 
 
 static void
@@ -17,11 +17,10 @@ alpha_test_quad(struct quad_stage *qs, struct quad_header *quad)
    const float ref = softpipe->depth_stencil->alpha.ref;
    unsigned passMask = 0x0, j;
    const uint cbuf = 0; /* only output[0].alpha is tested */
-   const float *aaaa = quad->outputs.color[cbuf][3];
+   const float *aaaa = quad->output.color[cbuf][3];
 
    switch (softpipe->depth_stencil->alpha.func) {
    case PIPE_FUNC_NEVER:
-      quad->mask = 0x0;
       break;
    case PIPE_FUNC_LESS:
       /*
@@ -76,9 +75,9 @@ alpha_test_quad(struct quad_stage *qs, struct quad_header *quad)
       assert(0);
    }
 
-   quad->mask &= passMask;
+   quad->inout.mask &= passMask;
 
-   if (quad->mask)
+   if (quad->inout.mask)
       qs->next->run(qs->next, quad);
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c
index 74c6bff84a..6f64c6e584 100644
--- a/src/gallium/drivers/softpipe/sp_quad_blend.c
+++ b/src/gallium/drivers/softpipe/sp_quad_blend.c
@@ -31,7 +31,8 @@
  */
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_headers.h"
 #include "sp_surface.h"
@@ -113,14 +114,14 @@ logicop_quad(struct quad_stage *qs, struct quad_header *quad)
       struct softpipe_cached_tile *
          tile = sp_get_cached_tile(softpipe,
                                    softpipe->cbuf_cache[cbuf],
-                                   quad->x0, quad->y0);
-      float (*quadColor)[4] = quad->outputs.color[cbuf];
+                                   quad->input.x0, quad->input.y0);
+      float (*quadColor)[4] = quad->output.color[cbuf];
       uint i, j;
 
       /* get/swizzle dest colors */
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1);
+         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
+         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
          for (i = 0; i < 4; i++) {
             dest[i][j] = tile->data.color[y][x][i];
          }
@@ -128,15 +129,15 @@ logicop_quad(struct quad_stage *qs, struct quad_header *quad)
 
       /* convert to ubyte */
       for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */
-         UNCLAMPED_FLOAT_TO_UBYTE(dst[j][0], dest[j][0]); /* P0 */
-         UNCLAMPED_FLOAT_TO_UBYTE(dst[j][1], dest[j][1]); /* P1 */
-         UNCLAMPED_FLOAT_TO_UBYTE(dst[j][2], dest[j][2]); /* P2 */
-         UNCLAMPED_FLOAT_TO_UBYTE(dst[j][3], dest[j][3]); /* P3 */
-
-         UNCLAMPED_FLOAT_TO_UBYTE(src[j][0], quadColor[j][0]); /* P0 */
-         UNCLAMPED_FLOAT_TO_UBYTE(src[j][1], quadColor[j][1]); /* P1 */
-         UNCLAMPED_FLOAT_TO_UBYTE(src[j][2], quadColor[j][2]); /* P2 */
-         UNCLAMPED_FLOAT_TO_UBYTE(src[j][3], quadColor[j][3]); /* P3 */
+         dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */
+         dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */
+         dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */
+         dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */
+
+         src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */
+         src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */
+         src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */
+         src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */
       }
 
       switch (softpipe->blend->logicop_func) {
@@ -209,10 +210,10 @@ logicop_quad(struct quad_stage *qs, struct quad_header *quad)
       }
 
       for (j = 0; j < 4; j++) {
-         quadColor[j][0] = UBYTE_TO_FLOAT(res[j][0]);
-         quadColor[j][1] = UBYTE_TO_FLOAT(res[j][1]);
-         quadColor[j][2] = UBYTE_TO_FLOAT(res[j][2]);
-         quadColor[j][3] = UBYTE_TO_FLOAT(res[j][3]);
+         quadColor[j][0] = ubyte_to_float(res[j][0]);
+         quadColor[j][1] = ubyte_to_float(res[j][1]);
+         quadColor[j][2] = ubyte_to_float(res[j][2]);
+         quadColor[j][3] = ubyte_to_float(res[j][3]);
       }
    }
 
@@ -243,14 +244,14 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
       struct softpipe_cached_tile *tile
          = sp_get_cached_tile(softpipe,
                               softpipe->cbuf_cache[cbuf],
-                              quad->x0, quad->y0);
-      float (*quadColor)[4] = quad->outputs.color[cbuf];
+                              quad->input.x0, quad->input.y0);
+      float (*quadColor)[4] = quad->output.color[cbuf];
       uint i, j;
 
       /* get/swizzle dest colors */
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1);
+         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
+         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
          for (i = 0; i < 4; i++) {
             dest[i][j] = tile->data.color[y][x][i];
          }
diff --git a/src/gallium/drivers/softpipe/sp_quad_bufloop.c b/src/gallium/drivers/softpipe/sp_quad_bufloop.c
index b3db428ef1..92e9af09c1 100644
--- a/src/gallium/drivers/softpipe/sp_quad_bufloop.c
+++ b/src/gallium/drivers/softpipe/sp_quad_bufloop.c
@@ -1,5 +1,5 @@
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_headers.h"
 #include "sp_surface.h"
diff --git a/src/gallium/drivers/softpipe/sp_quad_colormask.c b/src/gallium/drivers/softpipe/sp_quad_colormask.c
index 7fe080990b..f32bdfab78 100644
--- a/src/gallium/drivers/softpipe/sp_quad_colormask.c
+++ b/src/gallium/drivers/softpipe/sp_quad_colormask.c
@@ -31,7 +31,8 @@
  */
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_headers.h"
 #include "sp_surface.h"
@@ -55,14 +56,14 @@ colormask_quad(struct quad_stage *qs, struct quad_header *quad)
       struct softpipe_cached_tile *tile
          = sp_get_cached_tile(softpipe,
                               softpipe->cbuf_cache[cbuf],
-                              quad->x0, quad->y0);
-      float (*quadColor)[4] = quad->outputs.color[cbuf];
+                              quad->input.x0, quad->input.y0);
+      float (*quadColor)[4] = quad->output.color[cbuf];
       uint i, j;
 
       /* get/swizzle dest colors */
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->y0 & (TILE_SIZE-1)) + (j >> 1);
+         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
+         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
          for (i = 0; i < 4; i++) {
             dest[i][j] = tile->data.color[y][x][i];
          }
diff --git a/src/gallium/drivers/softpipe/sp_quad_coverage.c b/src/gallium/drivers/softpipe/sp_quad_coverage.c
index dd5ebb2296..ee29aa7dfe 100644
--- a/src/gallium/drivers/softpipe/sp_quad_coverage.c
+++ b/src/gallium/drivers/softpipe/sp_quad_coverage.c
@@ -33,7 +33,7 @@
 
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_headers.h"
 #include "sp_quad.h"
@@ -47,19 +47,19 @@ coverage_quad(struct quad_stage *qs, struct quad_header *quad)
 {
    struct softpipe_context *softpipe = qs->softpipe;
 
-   if ((softpipe->rasterizer->poly_smooth && quad->prim == PRIM_TRI) ||
-       (softpipe->rasterizer->line_smooth && quad->prim == PRIM_LINE) ||
-       (softpipe->rasterizer->point_smooth && quad->prim == PRIM_POINT)) {
+   if ((softpipe->rasterizer->poly_smooth && quad->input.prim == PRIM_TRI) ||
+       (softpipe->rasterizer->line_smooth && quad->input.prim == PRIM_LINE) ||
+       (softpipe->rasterizer->point_smooth && quad->input.prim == PRIM_POINT)) {
       uint cbuf;
 
       /* loop over colorbuffer outputs */
       for (cbuf = 0; cbuf < softpipe->framebuffer.num_cbufs; cbuf++) {
-         float (*quadColor)[4] = quad->outputs.color[cbuf];
+         float (*quadColor)[4] = quad->output.color[cbuf];
          unsigned j;
          for (j = 0; j < QUAD_SIZE; j++) {
-            assert(quad->coverage[j] >= 0.0);
-            assert(quad->coverage[j] <= 1.0);
-         quadColor[3][j] *= quad->coverage[j];
+            assert(quad->input.coverage[j] >= 0.0);
+            assert(quad->input.coverage[j] <= 1.0);
+         quadColor[3][j] *= quad->input.coverage[j];
          }
       }
    }
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
index 0c82692c6e..523bd3e080 100644
--- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -30,7 +30,7 @@
  */
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_headers.h"
 #include "sp_surface.h"
@@ -60,7 +60,7 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
    unsigned zmask = 0;
    unsigned j;
    struct softpipe_cached_tile *tile
-      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->x0, quad->y0);
+      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->input.x0, quad->input.y0);
 
    assert(ps); /* shouldn't get here if there's no zbuffer */
 
@@ -79,12 +79,12 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          float scale = 65535.0;
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             bzzzz[j] = tile->data.depth16[y][x];
          }
       }
@@ -94,12 +94,12 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          double scale = (double) (uint) ~0UL;
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             bzzzz[j] = tile->data.depth32[y][x];
          }
       }
@@ -111,12 +111,12 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          float scale = (float) ((1 << 24) - 1);
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             bzzzz[j] = tile->data.depth32[y][x] & 0xffffff;
          }
       }
@@ -128,12 +128,12 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          float scale = (float) ((1 << 24) - 1);
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            qzzzz[j] = (unsigned) (quad->outputs.depth[j] * scale);
+            qzzzz[j] = (unsigned) (quad->output.depth[j] * scale);
          }
 
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             bzzzz[j] = tile->data.depth32[y][x] >> 8;
          }
       }
@@ -192,14 +192,14 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
       assert(0);
    }
 
-   quad->mask &= zmask;
+   quad->inout.mask &= zmask;
 
    if (softpipe->depth_stencil->depth.writemask) {
       
       /* This is also efficient with sse / spe instructions: 
        */
       for (j = 0; j < QUAD_SIZE; j++) {
-	 if (quad->mask & (1 << j)) {
+	 if (quad->inout.mask & (1 << j)) {
 	    bzzzz[j] = qzzzz[j];
 	 }
       }
@@ -208,8 +208,8 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
       switch (format) {
       case PIPE_FORMAT_Z16_UNORM:
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             tile->data.depth16[y][x] = (ushort) bzzzz[j];
          }
          break;
@@ -218,15 +218,15 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          /* (yes, this falls through to a different case than above) */
       case PIPE_FORMAT_Z32_UNORM:
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             tile->data.depth32[y][x] = bzzzz[j];
          }
          break;
       case PIPE_FORMAT_S8Z24_UNORM:
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             uint s8z24 = tile->data.depth32[y][x];
             s8z24 = (s8z24 & 0xff000000) | bzzzz[j];
             tile->data.depth32[y][x] = s8z24;
@@ -234,8 +234,8 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          break;
       case PIPE_FORMAT_Z24S8_UNORM:
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             uint z24s8 = tile->data.depth32[y][x];
             z24s8 = (z24s8 & 0xff) | (bzzzz[j] << 8);
             tile->data.depth32[y][x] = z24s8;
@@ -243,8 +243,8 @@ sp_depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
          break;
       case PIPE_FORMAT_Z24X8_UNORM:
          for (j = 0; j < QUAD_SIZE; j++) {
-            int x = quad->x0 % TILE_SIZE + (j & 1);
-            int y = quad->y0 % TILE_SIZE + (j >> 1);
+            int x = quad->input.x0 % TILE_SIZE + (j & 1);
+            int y = quad->input.y0 % TILE_SIZE + (j >> 1);
             tile->data.depth32[y][x] = bzzzz[j] << 8;
          }
          break;
@@ -260,7 +260,7 @@ depth_test_quad(struct quad_stage *qs, struct quad_header *quad)
 {
    sp_depth_test_quad(qs, quad);
 
-   if (quad->mask)
+   if (quad->inout.mask)
       qs->next->run(qs->next, quad);
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_quad_earlyz.c b/src/gallium/drivers/softpipe/sp_quad_earlyz.c
index 22ea99049f..6e2dde304e 100644
--- a/src/gallium/drivers/softpipe/sp_quad_earlyz.c
+++ b/src/gallium/drivers/softpipe/sp_quad_earlyz.c
@@ -30,7 +30,7 @@
  */
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_headers.h"
 #include "sp_quad.h"
 
@@ -45,16 +45,16 @@ earlyz_quad(
    struct quad_stage    *qs,
    struct quad_header   *quad )
 {
-   const float fx = (float) quad->x0;
-   const float fy = (float) quad->y0;
+   const float fx = (float) quad->input.x0;
+   const float fy = (float) quad->input.y0;
    const float dzdx = quad->posCoef->dadx[2];
    const float dzdy = quad->posCoef->dady[2];
    const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy;
 
-   quad->outputs.depth[0] = z0;
-   quad->outputs.depth[1] = z0 + dzdx;
-   quad->outputs.depth[2] = z0 + dzdy;
-   quad->outputs.depth[3] = z0 + dzdx + dzdy;
+   quad->output.depth[0] = z0;
+   quad->output.depth[1] = z0 + dzdx;
+   quad->output.depth[2] = z0 + dzdy;
+   quad->output.depth[3] = z0 + dzdx + dzdy;
 
    qs->next->run( qs->next, quad );
 }
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index 8c88c192f8..1f0cb3e035 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -35,7 +35,8 @@
  * all the enabled attributes run contiguously.
  */
 
-#include "pipe/p_util.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_shader_tokens.h"
 
@@ -84,7 +85,7 @@ shade_quad(
    machine->InterpCoefs = quad->coef;
 
    /* run shader */
-   quad->mask &= softpipe->fs->run( softpipe->fs, 
+   quad->inout.mask &= softpipe->fs->run( softpipe->fs, 
 				    &qss->machine,
 				    quad );
 
@@ -100,16 +101,16 @@ shade_quad(
          case TGSI_SEMANTIC_COLOR:
             {
                uint cbuf = sem_index[i];
-               memcpy(quad->outputs.color[cbuf],
+               memcpy(quad->output.color[cbuf],
                       &machine->Outputs[i].xyzw[0].f[0],
-                      sizeof(quad->outputs.color[0]) );
+                      sizeof(quad->output.color[0]) );
             }
             break;
          case TGSI_SEMANTIC_POSITION:
             {
                uint j;
                for (j = 0; j < 4; j++) {
-                  quad->outputs.depth[j] = machine->Outputs[0].xyzw[2].f[j];
+                  quad->output.depth[j] = machine->Outputs[0].xyzw[2].f[j];
                }
                z_written = TRUE;
             }
@@ -121,20 +122,20 @@ shade_quad(
    if (!z_written) {
       /* compute Z values now, as in the quad earlyz stage */
       /* XXX we should really only do this if the earlyz stage is not used */
-      const float fx = (float) quad->x0;
-      const float fy = (float) quad->y0;
+      const float fx = (float) quad->input.x0;
+      const float fy = (float) quad->input.y0;
       const float dzdx = quad->posCoef->dadx[2];
       const float dzdy = quad->posCoef->dady[2];
       const float z0 = quad->posCoef->a0[2] + dzdx * fx + dzdy * fy;
 
-      quad->outputs.depth[0] = z0;
-      quad->outputs.depth[1] = z0 + dzdx;
-      quad->outputs.depth[2] = z0 + dzdy;
-      quad->outputs.depth[3] = z0 + dzdx + dzdy;
+      quad->output.depth[0] = z0;
+      quad->output.depth[1] = z0 + dzdx;
+      quad->output.depth[2] = z0 + dzdy;
+      quad->output.depth[3] = z0 + dzdx + dzdy;
    }
 
    /* shader may cull fragments */
-   if( quad->mask ) {
+   if( quad->inout.mask ) {
       qs->next->run( qs->next, quad );
    }
 }
diff --git a/src/gallium/drivers/softpipe/sp_quad_occlusion.c b/src/gallium/drivers/softpipe/sp_quad_occlusion.c
index 54254df1f1..169bd82876 100644
--- a/src/gallium/drivers/softpipe/sp_quad_occlusion.c
+++ b/src/gallium/drivers/softpipe/sp_quad_occlusion.c
@@ -33,7 +33,7 @@
 
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_headers.h"
 #include "sp_surface.h"
@@ -54,7 +54,7 @@ occlusion_count_quad(struct quad_stage *qs, struct quad_header *quad)
 {
    struct softpipe_context *softpipe = qs->softpipe;
 
-   softpipe->occlusion_count += count_bits(quad->mask);
+   softpipe->occlusion_count += count_bits(quad->inout.mask);
 
    qs->next->run(qs->next, quad);
 }
diff --git a/src/gallium/drivers/softpipe/sp_quad_output.c b/src/gallium/drivers/softpipe/sp_quad_output.c
index 40083138a4..d05e12d1d9 100644
--- a/src/gallium/drivers/softpipe/sp_quad_output.c
+++ b/src/gallium/drivers/softpipe/sp_quad_output.c
@@ -25,7 +25,7 @@
  * 
  **************************************************************************/
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_headers.h"
 #include "sp_surface.h"
@@ -41,8 +41,8 @@ static void
 output_quad(struct quad_stage *qs, struct quad_header *quad)
 {
    /* in-tile pos: */
-   const int itx = quad->x0 % TILE_SIZE;
-   const int ity = quad->y0 % TILE_SIZE;
+   const int itx = quad->input.x0 % TILE_SIZE;
+   const int ity = quad->input.y0 % TILE_SIZE;
 
    struct softpipe_context *softpipe = qs->softpipe;
    uint cbuf;
@@ -52,13 +52,13 @@ output_quad(struct quad_stage *qs, struct quad_header *quad)
       struct softpipe_cached_tile *tile
          = sp_get_cached_tile(softpipe,
                               softpipe->cbuf_cache[cbuf],
-                              quad->x0, quad->y0);
-      float (*quadColor)[4] = quad->outputs.color[cbuf];
+                              quad->input.x0, quad->input.y0);
+      float (*quadColor)[4] = quad->output.color[cbuf];
       int i, j;
 
       /* get/swizzle dest colors */
       for (j = 0; j < QUAD_SIZE; j++) {
-         if (quad->mask & (1 << j)) {
+         if (quad->inout.mask & (1 << j)) {
             int x = itx + (j & 1);
             int y = ity + (j >> 1);
             for (i = 0; i < 4; i++) { /* loop over color chans */
diff --git a/src/gallium/drivers/softpipe/sp_quad_stencil.c b/src/gallium/drivers/softpipe/sp_quad_stencil.c
index b4c7e942fa..abb5487748 100644
--- a/src/gallium/drivers/softpipe/sp_quad_stencil.c
+++ b/src/gallium/drivers/softpipe/sp_quad_stencil.c
@@ -10,7 +10,7 @@
 #include "sp_tile_cache.h"
 #include "sp_quad.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 
 
 /** Only 8-bit stencil supported */
@@ -206,9 +206,9 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
    ubyte ref, wrtMask, valMask;
    ubyte stencilVals[QUAD_SIZE];
    struct softpipe_cached_tile *tile
-      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->x0, quad->y0);
+      = sp_get_cached_tile(softpipe, softpipe->zsbuf_cache, quad->input.x0, quad->input.y0);
    uint j;
-   uint face = quad->facing;
+   uint face = quad->input.facing;
 
    if (!softpipe->depth_stencil->stencil[1].enabled) {
       /* single-sided stencil test, use front (face=0) state */
@@ -231,22 +231,22 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
    switch (ps->format) {
    case PIPE_FORMAT_S8Z24_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->x0 % TILE_SIZE + (j & 1);
-         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          stencilVals[j] = tile->data.depth32[y][x] >> 24;
       }
       break;
    case PIPE_FORMAT_Z24S8_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->x0 % TILE_SIZE + (j & 1);
-         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          stencilVals[j] = tile->data.depth32[y][x] & 0xff;
       }
       break;
    case PIPE_FORMAT_S8_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->x0 % TILE_SIZE + (j & 1);
-         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          stencilVals[j] = tile->data.stencil8[y][x];
       }
       break;
@@ -258,35 +258,35 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
    {
       unsigned passMask, failMask;
       passMask = do_stencil_test(stencilVals, func, ref, valMask);
-      failMask = quad->mask & ~passMask;
-      quad->mask &= passMask;
+      failMask = quad->inout.mask & ~passMask;
+      quad->inout.mask &= passMask;
 
       if (failOp != PIPE_STENCIL_OP_KEEP) {
          apply_stencil_op(stencilVals, failMask, failOp, ref, wrtMask);
       }
    }
 
-   if (quad->mask) {
+   if (quad->inout.mask) {
       /* now the pixels that passed the stencil test are depth tested */
       if (softpipe->depth_stencil->depth.enabled) {
-         const unsigned origMask = quad->mask;
+         const unsigned origMask = quad->inout.mask;
 
          sp_depth_test_quad(qs, quad);  /* quad->mask is updated */
 
          /* update stencil buffer values according to z pass/fail result */
          if (zFailOp != PIPE_STENCIL_OP_KEEP) {
-            const unsigned failMask = origMask & ~quad->mask;
+            const unsigned failMask = origMask & ~quad->inout.mask;
             apply_stencil_op(stencilVals, failMask, zFailOp, ref, wrtMask);
          }
 
          if (zPassOp != PIPE_STENCIL_OP_KEEP) {
-            const unsigned passMask = origMask & quad->mask;
+            const unsigned passMask = origMask & quad->inout.mask;
             apply_stencil_op(stencilVals, passMask, zPassOp, ref, wrtMask);
          }
       }
       else {
          /* no depth test, apply Zpass operator to stencil buffer values */
-         apply_stencil_op(stencilVals, quad->mask, zPassOp, ref, wrtMask);
+         apply_stencil_op(stencilVals, quad->inout.mask, zPassOp, ref, wrtMask);
       }
 
    }
@@ -295,8 +295,8 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
    switch (ps->format) {
    case PIPE_FORMAT_S8Z24_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->x0 % TILE_SIZE + (j & 1);
-         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          uint s8z24 = tile->data.depth32[y][x];
          s8z24 = (stencilVals[j] << 24) | (s8z24 & 0xffffff);
          tile->data.depth32[y][x] = s8z24;
@@ -304,8 +304,8 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
       break;
    case PIPE_FORMAT_Z24S8_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->x0 % TILE_SIZE + (j & 1);
-         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          uint z24s8 = tile->data.depth32[y][x];
          z24s8 = (z24s8 & 0xffffff00) | stencilVals[j];
          tile->data.depth32[y][x] = z24s8;
@@ -313,8 +313,8 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
       break;
    case PIPE_FORMAT_S8_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
-         int x = quad->x0 % TILE_SIZE + (j & 1);
-         int y = quad->y0 % TILE_SIZE + (j >> 1);
+         int x = quad->input.x0 % TILE_SIZE + (j & 1);
+         int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          tile->data.stencil8[y][x] = stencilVals[j];
       }
       break;
@@ -322,7 +322,7 @@ stencil_test_quad(struct quad_stage *qs, struct quad_header *quad)
       assert(0);
    }
 
-   if (quad->mask)
+   if (quad->inout.mask)
       qs->next->run(qs->next, quad);
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_quad_stipple.c b/src/gallium/drivers/softpipe/sp_quad_stipple.c
index f1e9b80e09..ccf37f6be5 100644
--- a/src/gallium/drivers/softpipe/sp_quad_stipple.c
+++ b/src/gallium/drivers/softpipe/sp_quad_stipple.c
@@ -7,7 +7,7 @@
 #include "sp_headers.h"
 #include "sp_quad.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 
 
 /**
@@ -19,17 +19,17 @@ stipple_quad(struct quad_stage *qs, struct quad_header *quad)
    static const uint bit31 = 1 << 31;
    static const uint bit30 = 1 << 30;
 
-   if (quad->prim == PRIM_TRI) {
+   if (quad->input.prim == PRIM_TRI) {
       struct softpipe_context *softpipe = qs->softpipe;
       /* need to invert Y to index into OpenGL's stipple pattern */
       int y0, y1;
       uint stipple0, stipple1;
       if (softpipe->rasterizer->origin_lower_left) {
-         y0 = softpipe->framebuffer.height - 1 - quad->y0;
+         y0 = softpipe->framebuffer.height - 1 - quad->input.y0;
          y1 = y0 - 1;
       }
       else {
-         y0 = quad->y0;
+         y0 = quad->input.y0;
          y1 = y0 + 1;
       }
       stipple0 = softpipe->poly_stipple.stipple[y0 % 32];
@@ -37,18 +37,18 @@ stipple_quad(struct quad_stage *qs, struct quad_header *quad)
 
 #if 1
       {
-      const int col0 = quad->x0 % 32;
+      const int col0 = quad->input.x0 % 32;
       if ((stipple0 & (bit31 >> col0)) == 0)
-         quad->mask &= ~MASK_TOP_LEFT;
+         quad->inout.mask &= ~MASK_TOP_LEFT;
 
       if ((stipple0 & (bit30 >> col0)) == 0)
-         quad->mask &= ~MASK_TOP_RIGHT;
+         quad->inout.mask &= ~MASK_TOP_RIGHT;
 
       if ((stipple1 & (bit31 >> col0)) == 0)
-         quad->mask &= ~MASK_BOTTOM_LEFT;
+         quad->inout.mask &= ~MASK_BOTTOM_LEFT;
 
       if ((stipple1 & (bit30 >> col0)) == 0)
-         quad->mask &= ~MASK_BOTTOM_RIGHT;
+         quad->inout.mask &= ~MASK_BOTTOM_RIGHT;
       }
 #else
       /* We'd like to use this code, but we'd need to redefine
@@ -56,11 +56,11 @@ stipple_quad(struct quad_stage *qs, struct quad_header *quad)
        * and similarly for the BOTTOM bits.  But that may have undesirable
        * side effects elsewhere.
        */
-      const int col0 = 30 - (quad->x0 % 32);
-      quad->mask &= (((stipple0 >> col0) & 0x3) | 
+      const int col0 = 30 - (quad->input.x0 % 32);
+      quad->inout.mask &= (((stipple0 >> col0) & 0x3) | 
                      (((stipple1 >> col0) & 0x3) << 2));
 #endif
-      if (!quad->mask)
+      if (!quad->inout.mask)
          return;
    }
 
diff --git a/src/gallium/drivers/softpipe/sp_query.c b/src/gallium/drivers/softpipe/sp_query.c
index adf9ccf64c..2106ee1d23 100644
--- a/src/gallium/drivers/softpipe/sp_query.c
+++ b/src/gallium/drivers/softpipe/sp_query.c
@@ -32,7 +32,7 @@
 #include "draw/draw_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_query.h"
 
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index f6b3d7ac24..9644dbd168 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -26,7 +26,7 @@
  **************************************************************************/
 
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_winsys.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index b7f2f16307..bc8263c33e 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -42,8 +42,11 @@
 #include "draw/draw_context.h"
 #include "draw/draw_private.h"
 #include "draw/draw_vertex.h"
-#include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
+#include "pipe/p_thread.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
 
 #define DEBUG_VERTS 0
 #define DEBUG_FRAGS 0
@@ -59,6 +62,87 @@ struct edge {
    int lines;		/**< number of lines on this edge */
 };
 
+#if SP_NUM_QUAD_THREADS > 1
+
+/* Set to 1 if you want other threads to be instantly
+ * notified of pending jobs.
+ */
+#define INSTANT_NOTEMPTY_NOTIFY 0
+
+struct thread_info
+{
+   struct setup_context *setup;
+   uint id;
+   pipe_thread handle;
+};
+
+struct quad_job;
+
+typedef void (* quad_job_routine)( struct setup_context *setup, uint thread, struct quad_job *job );
+
+struct quad_job
+{
+   struct quad_header_input input;
+   struct quad_header_inout inout;
+   quad_job_routine routine;
+};
+
+#define NUM_QUAD_JOBS 64
+
+struct quad_job_que
+{
+   struct quad_job jobs[NUM_QUAD_JOBS];
+   uint first;
+   uint last;
+   pipe_mutex que_mutex;
+   pipe_condvar que_notfull_condvar;
+   pipe_condvar que_notempty_condvar;
+   uint jobs_added;
+   uint jobs_done;
+   pipe_condvar que_done_condvar;
+};
+
+static void
+add_quad_job( struct quad_job_que *que, struct quad_header *quad, quad_job_routine routine )
+{
+#if INSTANT_NOTEMPTY_NOTIFY
+   boolean empty;
+#endif
+
+   /* Wait for empty slot, see if the que is empty.
+    */
+   pipe_mutex_lock( que->que_mutex );
+   while ((que->last + 1) % NUM_QUAD_JOBS == que->first) {
+#if !INSTANT_NOTEMPTY_NOTIFY
+      pipe_condvar_broadcast( que->que_notempty_condvar );
+#endif
+      pipe_condvar_wait( que->que_notfull_condvar, que->que_mutex );
+   }
+#if INSTANT_NOTEMPTY_NOTIFY
+   empty = que->last == que->first;
+#endif
+   que->jobs_added++;
+   pipe_mutex_unlock( que->que_mutex );
+
+   /* Submit new job.
+    */
+   que->jobs[que->last].input = quad->input;
+   que->jobs[que->last].inout = quad->inout;
+   que->jobs[que->last].routine = routine;
+   que->last = (que->last + 1) % NUM_QUAD_JOBS;
+
+#if INSTANT_NOTEMPTY_NOTIFY
+   /* If the que was empty, notify consumers there's a job to be done.
+    */
+   if (empty) {
+      pipe_mutex_lock( que->que_mutex );
+      pipe_condvar_broadcast( que->que_notempty_condvar );
+      pipe_mutex_unlock( que->que_mutex );
+   }
+#endif
+}
+
+#endif
 
 /**
  * Triangle setup info (derived from draw_stage).
@@ -86,6 +170,11 @@ struct setup_context {
    struct tgsi_interp_coef posCoef;  /* For Z, W */
    struct quad_header quad;
 
+#if SP_NUM_QUAD_THREADS > 1
+   struct quad_job_que que;
+   struct thread_info threads[SP_NUM_QUAD_THREADS];
+#endif
+
    struct {
       int left[2];   /**< [0] = row0, [1] = row1 */
       int right[2];
@@ -102,8 +191,78 @@ struct setup_context {
    unsigned winding;		/* which winding to cull */
 };
 
+#if SP_NUM_QUAD_THREADS > 1
+
+static PIPE_THREAD_ROUTINE( quad_thread, param )
+{
+   struct thread_info *info = (struct thread_info *) param;
+   struct quad_job_que *que = &info->setup->que;
+
+   for (;;) {
+      struct quad_job job;
+      boolean full;
+
+      /* Wait for an available job.
+       */
+      pipe_mutex_lock( que->que_mutex );
+      while (que->last == que->first)
+         pipe_condvar_wait( que->que_notempty_condvar, que->que_mutex );
+
+      /* See if the que is full.
+       */
+      full = (que->last + 1) % NUM_QUAD_JOBS == que->first;
+
+      /* Take a job and remove it from que.
+       */
+      job = que->jobs[que->first];
+      que->first = (que->first + 1) % NUM_QUAD_JOBS;
+
+      /* Notify the producer if the que is not full.
+       */
+      if (full)
+         pipe_condvar_signal( que->que_notfull_condvar );
+      pipe_mutex_unlock( que->que_mutex );
+
+      job.routine( info->setup, info->id, &job );
+
+      /* Notify the producer if that's the last finished job.
+       */
+      pipe_mutex_lock( que->que_mutex );
+      que->jobs_done++;
+      if (que->jobs_added == que->jobs_done)
+         pipe_condvar_signal( que->que_done_condvar );
+      pipe_mutex_unlock( que->que_mutex );
+   }
 
+   return NULL;
+}
+
+#define WAIT_FOR_COMPLETION(setup) \
+   do {\
+      pipe_mutex_lock( setup->que.que_mutex );\
+      if (!INSTANT_NOTEMPTY_NOTIFY)\
+         pipe_condvar_broadcast( setup->que.que_notempty_condvar );\
+      while (setup->que.jobs_added != setup->que.jobs_done)\
+         pipe_condvar_wait( setup->que.que_done_condvar, setup->que.que_mutex );\
+      pipe_mutex_unlock( setup->que.que_mutex );\
+   } while (0)
 
+#else
+
+#define WAIT_FOR_COMPLETION(setup) ((void) 0)
+
+#endif
+
+/**
+ * Test if x is NaN or +/- infinity.
+ */
+static INLINE boolean
+is_inf_or_nan(float x)
+{
+   union fi tmp;
+   tmp.f = x;
+   return !(int)((unsigned int)((tmp.i & 0x7fffffff)-0x7f800000) >> 31);
+}
 
 
 static boolean cull_tri( struct setup_context *setup,
@@ -131,7 +290,7 @@ static boolean cull_tri( struct setup_context *setup,
  * Clip setup->quad against the scissor/surface bounds.
  */
 static INLINE void
-quad_clip(struct setup_context *setup)
+quad_clip( struct setup_context *setup, struct quad_header *quad )
 {
    const struct pipe_scissor_state *cliprect = &setup->softpipe->cliprect;
    const int minx = (int) cliprect->minx;
@@ -139,22 +298,22 @@ quad_clip(struct setup_context *setup)
    const int miny = (int) cliprect->miny;
    const int maxy = (int) cliprect->maxy;
 
-   if (setup->quad.x0 >= maxx ||
-       setup->quad.y0 >= maxy ||
-       setup->quad.x0 + 1 < minx ||
-       setup->quad.y0 + 1 < miny) {
+   if (quad->input.x0 >= maxx ||
+       quad->input.y0 >= maxy ||
+       quad->input.x0 + 1 < minx ||
+       quad->input.y0 + 1 < miny) {
       /* totally clipped */
-      setup->quad.mask = 0x0;
+      quad->inout.mask = 0x0;
       return;
    }
-   if (setup->quad.x0 < minx)
-      setup->quad.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
-   if (setup->quad.y0 < miny)
-      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
-   if (setup->quad.x0 == maxx - 1)
-      setup->quad.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
-   if (setup->quad.y0 == maxy - 1)
-      setup->quad.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
+   if (quad->input.x0 < minx)
+      quad->inout.mask &= (MASK_BOTTOM_RIGHT | MASK_TOP_RIGHT);
+   if (quad->input.y0 < miny)
+      quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_BOTTOM_RIGHT);
+   if (quad->input.x0 == maxx - 1)
+      quad->inout.mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
+   if (quad->input.y0 == maxy - 1)
+      quad->inout.mask &= (MASK_TOP_LEFT | MASK_TOP_RIGHT);
 }
 
 
@@ -162,35 +321,59 @@ quad_clip(struct setup_context *setup)
  * Emit a quad (pass to next stage) with clipping.
  */
 static INLINE void
-clip_emit_quad(struct setup_context *setup)
+clip_emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread )
 {
-   quad_clip(setup);
-   if (setup->quad.mask) {
+   quad_clip( setup, quad );
+   if (quad->inout.mask) {
       struct softpipe_context *sp = setup->softpipe;
-      sp->quad.first->run(sp->quad.first, &setup->quad);
+
+      sp->quad[thread].first->run( sp->quad[thread].first, quad );
    }
 }
 
+#if SP_NUM_QUAD_THREADS > 1
+
+static void
+clip_emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
+{
+   struct quad_header quad;
+
+   quad.input = job->input;
+   quad.inout = job->inout;
+   quad.coef = setup->quad.coef;
+   quad.posCoef = setup->quad.posCoef;
+   quad.nr_attrs = setup->quad.nr_attrs;
+   clip_emit_quad( setup, &quad, thread );
+}
+
+#define CLIP_EMIT_QUAD(setup) add_quad_job( &setup->que, &setup->quad, clip_emit_quad_job )
+
+#else
+
+#define CLIP_EMIT_QUAD(setup) clip_emit_quad( setup, &setup->quad, 0 )
+
+#endif
 
 /**
  * Emit a quad (pass to next stage).  No clipping is done.
  */
 static INLINE void
-emit_quad( struct setup_context *setup, int x, int y, unsigned mask )
+emit_quad( struct setup_context *setup, struct quad_header *quad, uint thread )
 {
    struct softpipe_context *sp = setup->softpipe;
-   setup->quad.x0 = x;
-   setup->quad.y0 = y;
-   setup->quad.mask = mask;
+#if DEBUG_FRAGS
+   uint mask = quad->inout.mask;
+#endif
+
 #if DEBUG_FRAGS
    if (mask & 1) setup->numFragsEmitted++;
    if (mask & 2) setup->numFragsEmitted++;
    if (mask & 4) setup->numFragsEmitted++;
    if (mask & 8) setup->numFragsEmitted++;
 #endif
-   sp->quad.first->run(sp->quad.first, &setup->quad);
+   sp->quad[thread].first->run( sp->quad[thread].first, quad );
 #if DEBUG_FRAGS
-   mask = setup->quad.mask;
+   mask = quad->inout.mask;
    if (mask & 1) setup->numFragsWritten++;
    if (mask & 2) setup->numFragsWritten++;
    if (mask & 4) setup->numFragsWritten++;
@@ -198,6 +381,38 @@ emit_quad( struct setup_context *setup, int x, int y, unsigned mask )
 #endif
 }
 
+#if SP_NUM_QUAD_THREADS > 1
+
+static void
+emit_quad_job( struct setup_context *setup, uint thread, struct quad_job *job )
+{
+   struct quad_header quad;
+
+   quad.input = job->input;
+   quad.inout = job->inout;
+   quad.coef = setup->quad.coef;
+   quad.posCoef = setup->quad.posCoef;
+   quad.nr_attrs = setup->quad.nr_attrs;
+   emit_quad( setup, &quad, thread );
+}
+
+#define EMIT_QUAD(setup,x,y,mask) do {\
+      setup->quad.input.x0 = x;\
+      setup->quad.input.y0 = y;\
+      setup->quad.inout.mask = mask;\
+      add_quad_job( &setup->que, &setup->quad, emit_quad_job );\
+   } while (0)
+
+#else
+
+#define EMIT_QUAD(setup,x,y,mask) do {\
+      setup->quad.input.x0 = x;\
+      setup->quad.input.y0 = y;\
+      setup->quad.inout.mask = mask;\
+      emit_quad( setup, &setup->quad, 0 );\
+   } while (0)
+
+#endif
 
 /**
  * Given an X or Y coordinate, return the block/quad coordinate that it
@@ -237,7 +452,7 @@ static void flush_spans( struct setup_context *setup )
             mask |= MASK_TOP_RIGHT;
          if (x+1 >= xleft1 && x+1 < xright1)
             mask |= MASK_BOTTOM_RIGHT;
-         emit_quad( setup, x, setup->span.y, mask );
+         EMIT_QUAD( setup, x, setup->span.y, mask );
       }
       break;
 
@@ -251,7 +466,7 @@ static void flush_spans( struct setup_context *setup )
             mask |= MASK_TOP_LEFT;
          if (x+1 >= xleft0 && x+1 < xright0)
             mask |= MASK_TOP_RIGHT;
-         emit_quad( setup, x, setup->span.y, mask );
+         EMIT_QUAD( setup, x, setup->span.y, mask );
       }
       break;
 
@@ -265,7 +480,7 @@ static void flush_spans( struct setup_context *setup )
             mask |= MASK_BOTTOM_LEFT;
          if (x+1 >= xleft1 && x+1 < xright1)
             mask |= MASK_BOTTOM_RIGHT;
-         emit_quad( setup, x, setup->span.y, mask );
+         EMIT_QUAD( setup, x, setup->span.y, mask );
       }
       break;
 
@@ -293,6 +508,9 @@ static void print_vertex(const struct setup_context *setup,
 }
 #endif
 
+/**
+ * \return FALSE if coords are inf/nan (cull the tri), TRUE otherwise
+ */
 static boolean setup_sort_vertices( struct setup_context *setup,
                                     float det,
                                     const float (*v0)[4],
@@ -370,17 +588,20 @@ static boolean setup_sort_vertices( struct setup_context *setup,
 			    setup->ebot.dx * setup->emaj.dy);
 
       setup->oneoverarea = 1.0f / area;
+
       /*
       debug_printf("%s one-over-area %f  area %f  det %f\n",
                    __FUNCTION__, setup->oneoverarea, area, det );
       */
+      if (is_inf_or_nan(setup->oneoverarea))
+         return FALSE;
    }
 
    /* We need to know if this is a front or back-facing triangle for:
     *  - the GLSL gl_FrontFacing fragment attribute (bool)
     *  - two-sided stencil test
     */
-   setup->quad.facing = (det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
+   setup->quad.input.facing = (det > 0.0) ^ (setup->softpipe->rasterizer->front_winding == PIPE_WINDING_CW);
 
    return TRUE;
 }
@@ -577,7 +798,7 @@ static void setup_tri_coefficients( struct setup_context *setup )
 
       if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
          /* FOG.y = front/back facing  XXX fix this */
-         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.input.facing;
          setup->coef[fragSlot].dadx[1] = 0.0;
          setup->coef[fragSlot].dady[1] = 0.0;
       }
@@ -595,18 +816,18 @@ static void setup_tri_edges( struct setup_context *setup )
    float vmid_y = setup->vmid[0][1] - 0.5f;
    float vmax_y = setup->vmax[0][1] - 0.5f;
 
-   setup->emaj.sy = CEILF(vmin_y);
-   setup->emaj.lines = (int) CEILF(vmax_y - setup->emaj.sy);
+   setup->emaj.sy = ceilf(vmin_y);
+   setup->emaj.lines = (int) ceilf(vmax_y - setup->emaj.sy);
    setup->emaj.dxdy = setup->emaj.dx / setup->emaj.dy;
    setup->emaj.sx = vmin_x + (setup->emaj.sy - vmin_y) * setup->emaj.dxdy;
 
-   setup->etop.sy = CEILF(vmid_y);
-   setup->etop.lines = (int) CEILF(vmax_y - setup->etop.sy);
+   setup->etop.sy = ceilf(vmid_y);
+   setup->etop.lines = (int) ceilf(vmax_y - setup->etop.sy);
    setup->etop.dxdy = setup->etop.dx / setup->etop.dy;
    setup->etop.sx = vmid_x + (setup->etop.sy - vmid_y) * setup->etop.dxdy;
 
-   setup->ebot.sy = CEILF(vmin_y);
-   setup->ebot.lines = (int) CEILF(vmid_y - setup->ebot.sy);
+   setup->ebot.sy = ceilf(vmin_y);
+   setup->ebot.lines = (int) ceilf(vmid_y - setup->ebot.sy);
    setup->ebot.dxdy = setup->ebot.dx / setup->ebot.dy;
    setup->ebot.sx = vmin_x + (setup->ebot.sy - vmin_y) * setup->ebot.dxdy;
 }
@@ -742,11 +963,12 @@ void setup_tri( struct setup_context *setup,
    if (cull_tri( setup, det ))
       return;
 
-   setup_sort_vertices( setup, det, v0, v1, v2 );
+   if (!setup_sort_vertices( setup, det, v0, v1, v2 ))
+      return;
    setup_tri_coefficients( setup );
    setup_tri_edges( setup );
 
-   setup->quad.prim = PRIM_TRI;
+   setup->quad.input.prim = PRIM_TRI;
 
    setup->span.y = 0;
    setup->span.y_flags = 0;
@@ -771,6 +993,8 @@ void setup_tri( struct setup_context *setup,
 
    flush_spans( setup );
 
+   WAIT_FOR_COMPLETION(setup);
+
 #if DEBUG_FRAGS
    printf("Tri: %u frags emitted, %u written\n",
           setup->numFragsEmitted,
@@ -827,7 +1051,7 @@ line_persp_coeff(struct setup_context *setup,
  * Compute the setup->coef[] array dadx, dady, a0 values.
  * Must be called after setup->vmin,vmax are initialized.
  */
-static INLINE void
+static INLINE boolean
 setup_line_coefficients(struct setup_context *setup,
                         const float (*v0)[4],
                         const float (*v1)[4])
@@ -836,6 +1060,7 @@ setup_line_coefficients(struct setup_context *setup,
    const struct sp_fragment_shader *spfs = softpipe->fs;
    const struct vertex_info *vinfo = softpipe_get_vertex_info(softpipe);
    uint fragSlot;
+   float area;
 
    /* use setup->vmin, vmax to point to vertices */
    setup->vprovoke = v1;
@@ -844,9 +1069,12 @@ setup_line_coefficients(struct setup_context *setup,
 
    setup->emaj.dx = setup->vmax[0][0] - setup->vmin[0][0];
    setup->emaj.dy = setup->vmax[0][1] - setup->vmin[0][1];
-   /* NOTE: this is not really 1/area */
-   setup->oneoverarea = 1.0f / (setup->emaj.dx * setup->emaj.dx +
-                                setup->emaj.dy * setup->emaj.dy);
+
+   /* NOTE: this is not really area but something proportional to it */
+   area = setup->emaj.dx * setup->emaj.dx + setup->emaj.dy * setup->emaj.dy;
+   if (area == 0.0f || is_inf_or_nan(area))
+      return FALSE;
+   setup->oneoverarea = 1.0f / area;
 
    /* z and w are done by linear interpolation:
     */
@@ -881,11 +1109,12 @@ setup_line_coefficients(struct setup_context *setup,
 
       if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
          /* FOG.y = front/back facing  XXX fix this */
-         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.input.facing;
          setup->coef[fragSlot].dadx[1] = 0.0;
          setup->coef[fragSlot].dady[1] = 0.0;
       }
    }
+   return TRUE;
 }
 
 
@@ -901,20 +1130,20 @@ plot(struct setup_context *setup, int x, int y)
    const int quadY = y - iy;
    const int mask = (1 << ix) << (2 * iy);
 
-   if (quadX != setup->quad.x0 ||
-       quadY != setup->quad.y0)
+   if (quadX != setup->quad.input.x0 ||
+       quadY != setup->quad.input.y0)
    {
       /* flush prev quad, start new quad */
 
-      if (setup->quad.x0 != -1)
-         clip_emit_quad(setup);
+      if (setup->quad.input.x0 != -1)
+         CLIP_EMIT_QUAD(setup);
 
-      setup->quad.x0 = quadX;
-      setup->quad.y0 = quadY;
-      setup->quad.mask = 0x0;
+      setup->quad.input.x0 = quadX;
+      setup->quad.input.y0 = quadY;
+      setup->quad.inout.mask = 0x0;
    }
 
-   setup->quad.mask |= mask;
+   setup->quad.inout.mask |= mask;
 }
 
 
@@ -942,18 +1171,19 @@ setup_line(struct setup_context *setup,
    print_vertex(setup, v1);
 #endif
 
-   assert(v0[0][0] < 1.0e9);
-   assert(v0[0][1] < 1.0e9);
-   assert(v1[0][0] < 1.0e9);
-   assert(v1[0][1] < 1.0e9);
-
    if (setup->softpipe->no_rast)
       return;
 
    if (dx == 0 && dy == 0)
       return;
 
-   setup_line_coefficients(setup, v0, v1);
+   if (!setup_line_coefficients(setup, v0, v1))
+      return;
+
+   assert(v0[0][0] < 1.0e9);
+   assert(v0[0][1] < 1.0e9);
+   assert(v1[0][0] < 1.0e9);
+   assert(v1[0][1] < 1.0e9);
 
    if (dx < 0) {
       dx = -dx;   /* make positive */
@@ -974,16 +1204,16 @@ setup_line(struct setup_context *setup,
    assert(dx >= 0);
    assert(dy >= 0);
 
-   setup->quad.x0 = setup->quad.y0 = -1;
-   setup->quad.mask = 0x0;
-   setup->quad.prim = PRIM_LINE;
+   setup->quad.input.x0 = setup->quad.input.y0 = -1;
+   setup->quad.inout.mask = 0x0;
+   setup->quad.input.prim = PRIM_LINE;
    /* XXX temporary: set coverage to 1.0 so the line appears
     * if AA mode happens to be enabled.
     */
-   setup->quad.coverage[0] =
-   setup->quad.coverage[1] =
-   setup->quad.coverage[2] =
-   setup->quad.coverage[3] = 1.0;
+   setup->quad.input.coverage[0] =
+   setup->quad.input.coverage[1] =
+   setup->quad.input.coverage[2] =
+   setup->quad.input.coverage[3] = 1.0;
 
    if (dx > dy) {
       /*** X-major line ***/
@@ -1027,9 +1257,11 @@ setup_line(struct setup_context *setup,
    }
 
    /* draw final quad */
-   if (setup->quad.mask) {
-      clip_emit_quad(setup);
+   if (setup->quad.inout.mask) {
+      CLIP_EMIT_QUAD(setup);
    }
+
+   WAIT_FOR_COMPLETION(setup);
 }
 
 
@@ -1123,22 +1355,22 @@ setup_point( struct setup_context *setup,
 
       if (spfs->info.input_semantic_name[fragSlot] == TGSI_SEMANTIC_FOG) {
          /* FOG.y = front/back facing  XXX fix this */
-         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.facing;
+         setup->coef[fragSlot].a0[1] = 1.0f - setup->quad.input.facing;
          setup->coef[fragSlot].dadx[1] = 0.0;
          setup->coef[fragSlot].dady[1] = 0.0;
       }
    }
 
-   setup->quad.prim = PRIM_POINT;
+   setup->quad.input.prim = PRIM_POINT;
 
    if (halfSize <= 0.5 && !round) {
       /* special case for 1-pixel points */
       const int ix = ((int) x) & 1;
       const int iy = ((int) y) & 1;
-      setup->quad.x0 = (int) x - ix;
-      setup->quad.y0 = (int) y - iy;
-      setup->quad.mask = (1 << ix) << (2 * iy);
-      clip_emit_quad(setup);
+      setup->quad.input.x0 = (int) x - ix;
+      setup->quad.input.y0 = (int) y - iy;
+      setup->quad.inout.mask = (1 << ix) << (2 * iy);
+      CLIP_EMIT_QUAD(setup);
    }
    else {
       if (round) {
@@ -1158,15 +1390,15 @@ setup_point( struct setup_context *setup,
             for (ix = ixmin; ix <= ixmax; ix += 2) {
                float dx, dy, dist2, cover;
 
-               setup->quad.mask = 0x0;
+               setup->quad.inout.mask = 0x0;
 
                dx = (ix + 0.5f) - x;
                dy = (iy + 0.5f) - y;
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_TOP_LEFT;
+                  setup->quad.input.coverage[QUAD_TOP_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad.inout.mask |= MASK_TOP_LEFT;
                }
 
                dx = (ix + 1.5f) - x;
@@ -1174,8 +1406,8 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_TOP_RIGHT;
+                  setup->quad.input.coverage[QUAD_TOP_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad.inout.mask |= MASK_TOP_RIGHT;
                }
 
                dx = (ix + 0.5f) - x;
@@ -1183,8 +1415,8 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_BOTTOM_LEFT;
+                  setup->quad.input.coverage[QUAD_BOTTOM_LEFT] = MIN2(cover, 1.0f);
+                  setup->quad.inout.mask |= MASK_BOTTOM_LEFT;
                }
 
                dx = (ix + 1.5f) - x;
@@ -1192,14 +1424,14 @@ setup_point( struct setup_context *setup,
                dist2 = dx * dx + dy * dy;
                if (dist2 <= rmax2) {
                   cover = 1.0F - (dist2 - rmin2) * cscale;
-                  setup->quad.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
-                  setup->quad.mask |= MASK_BOTTOM_RIGHT;
+                  setup->quad.input.coverage[QUAD_BOTTOM_RIGHT] = MIN2(cover, 1.0f);
+                  setup->quad.inout.mask |= MASK_BOTTOM_RIGHT;
                }
 
-               if (setup->quad.mask) {
-                  setup->quad.x0 = ix;
-                  setup->quad.y0 = iy;
-                  clip_emit_quad(setup);
+               if (setup->quad.inout.mask) {
+                  setup->quad.input.x0 = ix;
+                  setup->quad.input.y0 = iy;
+                  CLIP_EMIT_QUAD(setup);
                }
             }
          }
@@ -1243,14 +1475,16 @@ setup_point( struct setup_context *setup,
                   mask &= (MASK_BOTTOM_LEFT | MASK_TOP_LEFT);
                }
 
-               setup->quad.mask = mask;
-               setup->quad.x0 = ix;
-               setup->quad.y0 = iy;
-               clip_emit_quad(setup);
+               setup->quad.inout.mask = mask;
+               setup->quad.input.x0 = ix;
+               setup->quad.input.y0 = iy;
+               CLIP_EMIT_QUAD(setup);
             }
          }
       }
    }
+
+   WAIT_FOR_COMPLETION(setup);
 }
 
 void setup_prepare( struct setup_context *setup )
@@ -1275,7 +1509,9 @@ void setup_prepare( struct setup_context *setup )
    /* Note: nr_attrs is only used for debugging (vertex printing) */
    setup->quad.nr_attrs = draw_num_vs_outputs(sp->draw);
 
-   sp->quad.first->begin(sp->quad.first);
+   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
+      sp->quad[i].first->begin( sp->quad[i].first );
+   }
 
    if (sp->reduced_api_prim == PIPE_PRIM_TRIANGLES &&
        sp->rasterizer->fill_cw == PIPE_POLYGON_MODE_FILL &&
@@ -1303,11 +1539,31 @@ void setup_destroy_context( struct setup_context *setup )
 struct setup_context *setup_create_context( struct softpipe_context *softpipe )
 {
    struct setup_context *setup = CALLOC_STRUCT(setup_context);
+#if SP_NUM_QUAD_THREADS > 1
+   uint i;
+#endif
 
    setup->softpipe = softpipe;
 
    setup->quad.coef = setup->coef;
    setup->quad.posCoef = &setup->posCoef;
 
+#if SP_NUM_QUAD_THREADS > 1
+   setup->que.first = 0;
+   setup->que.last = 0;
+   pipe_mutex_init( setup->que.que_mutex );
+   pipe_condvar_init( setup->que.que_notfull_condvar );
+   pipe_condvar_init( setup->que.que_notempty_condvar );
+   setup->que.jobs_added = 0;
+   setup->que.jobs_done = 0;
+   pipe_condvar_init( setup->que.que_done_condvar );
+   for (i = 0; i < SP_NUM_QUAD_THREADS; i++) {
+      setup->threads[i].setup = setup;
+      setup->threads[i].id = i;
+      setup->threads[i].handle = pipe_thread_create( quad_thread, &setup->threads[i] );
+   }
+#endif
+
    return setup;
 }
+
diff --git a/src/gallium/drivers/softpipe/sp_state_blend.c b/src/gallium/drivers/softpipe/sp_state_blend.c
index 2d40d6bd8f..384fe559af 100644
--- a/src/gallium/drivers/softpipe/sp_state_blend.c
+++ b/src/gallium/drivers/softpipe/sp_state_blend.c
@@ -28,7 +28,7 @@
 /* Authors:  Keith Whitwell <keith@tungstengraphics.com>
  */
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_state.h"
 
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
index f10a1fa471..6b6a4c3ff3 100644
--- a/src/gallium/drivers/softpipe/sp_state_derived.c
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -25,7 +25,8 @@
  * 
  **************************************************************************/
 
-#include "pipe/p_util.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 #include "pipe/p_shader_tokens.h"
 #include "draw/draw_context.h"
 #include "draw/draw_vertex.h"
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index 76fe6bfef9..e5b609cf6c 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -30,7 +30,7 @@
 #include "sp_fs.h"
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_inlines.h"
 #include "pipe/p_winsys.h"
 #include "pipe/p_shader_tokens.h"
@@ -152,7 +152,7 @@ softpipe_set_constant_buffer(struct pipe_context *pipe,
    assert(index == 0);
 
    /* note: reference counting */
-   pipe_buffer_reference(ws,
+   winsys_buffer_reference(ws,
 			 &softpipe->constants[shader].buffer,
 			 buf ? buf->buffer : NULL);
    softpipe->constants[shader].size = buf ? buf->size : 0;
diff --git a/src/gallium/drivers/softpipe/sp_state_rasterizer.c b/src/gallium/drivers/softpipe/sp_state_rasterizer.c
index 98e04352db..87b7219683 100644
--- a/src/gallium/drivers/softpipe/sp_state_rasterizer.c
+++ b/src/gallium/drivers/softpipe/sp_state_rasterizer.c
@@ -26,7 +26,7 @@
  **************************************************************************/
 
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "sp_context.h"
 #include "sp_state.h"
 #include "draw/draw_context.h"
diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c
index 033288a0aa..99a28c0d7e 100644
--- a/src/gallium/drivers/softpipe/sp_state_sampler.c
+++ b/src/gallium/drivers/softpipe/sp_state_sampler.c
@@ -29,7 +29,7 @@
  *  Brian Paul
  */
 
-#include "pipe/p_util.h"
+#include "util/u_memory.h"
 #include "pipe/p_inlines.h"
 
 #include "draw/draw_context.h"
diff --git a/src/gallium/drivers/softpipe/sp_surface.c b/src/gallium/drivers/softpipe/sp_surface.c
index 7dc15c38d1..6ade732698 100644
--- a/src/gallium/drivers/softpipe/sp_surface.c
+++ b/src/gallium/drivers/softpipe/sp_surface.c
@@ -25,129 +25,14 @@
  * 
  **************************************************************************/
 
-#include "pipe/p_defines.h"
-#include "pipe/p_util.h"
-#include "pipe/p_inlines.h"
-#include "pipe/p_winsys.h"
-#include "util/p_tile.h"
+#include "util/u_rect.h"
 #include "sp_context.h"
-#include "sp_surface.h"
 
 
 
-/**
- * Copy a rectangular region from one surface to another.
- * Surfaces must have same bpp.
- *
- * Assumes all values are within bounds -- no checking at this level -
- * do it higher up if required.
- */
-static void
-sp_surface_copy(struct pipe_context *pipe,
-                boolean do_flip,
-		struct pipe_surface *dst,
-		unsigned dstx, unsigned dsty,
-		struct pipe_surface *src,
-		unsigned srcx, unsigned srcy, unsigned width, unsigned height)
-{
-   void *dst_map = pipe->screen->surface_map( pipe->screen,
-                                              dst,
-                                              PIPE_BUFFER_USAGE_CPU_WRITE );
-
-   const void *src_map = pipe->screen->surface_map( pipe->screen,
-                                                    src,
-                                                    PIPE_BUFFER_USAGE_CPU_READ );
-
-   assert(dst->block.size == src->block.size);
-   assert(dst->block.width == src->block.width);
-   assert(dst->block.height == src->block.height);
-   assert(src_map);
-   assert(dst_map);
-
-   /* If do_flip, invert src_y position and pass negative src stride */
-   pipe_copy_rect(dst_map,
-                  &dst->block,
-                  dst->stride,
-                  dstx, dsty,
-                  width, height,
-                  src_map,
-                  do_flip ? -(int) src->stride : src->stride,
-                  srcx, do_flip ? src->height - 1 - srcy : srcy);
-
-   pipe->screen->surface_unmap(pipe->screen, src);
-   pipe->screen->surface_unmap(pipe->screen, dst);
-}
-
-
-static void *
-get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y)
-{
-   return (char *)dst_map + y / dst->block.height * dst->stride + x / dst->block.width * dst->block.size;
-}
-
-
-#define UBYTE_TO_USHORT(B) ((B) | ((B) << 8))
-
-
-/**
- * Fill a rectangular sub-region.  Need better logic about when to
- * push buffers into AGP - will currently do so whenever possible.
- */
-static void
-sp_surface_fill(struct pipe_context *pipe,
-		struct pipe_surface *dst,
-		unsigned dstx, unsigned dsty,
-		unsigned width, unsigned height, unsigned value)
-{
-   unsigned i, j;
-   void *dst_map = pipe->screen->surface_map( pipe->screen,
-                                              dst,
-                                              PIPE_BUFFER_USAGE_CPU_WRITE );
-
-   assert(dst->stride > 0);
-
-
-   switch (dst->block.size) {
-   case 1:
-   case 2:
-   case 4:
-      pipe_fill_rect(dst_map, &dst->block, dst->stride, dstx, dsty, width, height, value);
-      break;
-   case 8:
-      {
-         /* expand the 4-byte clear value to an 8-byte value */
-         ushort *row = (ushort *) get_pointer(dst, dst_map, dstx, dsty);
-         ushort val0 = UBYTE_TO_USHORT((value >>  0) & 0xff);
-         ushort val1 = UBYTE_TO_USHORT((value >>  8) & 0xff);
-         ushort val2 = UBYTE_TO_USHORT((value >> 16) & 0xff);
-         ushort val3 = UBYTE_TO_USHORT((value >> 24) & 0xff);
-         val0 = (val0 << 8) | val0;
-         val1 = (val1 << 8) | val1;
-         val2 = (val2 << 8) | val2;
-         val3 = (val3 << 8) | val3;
-         for (i = 0; i < height; i++) {
-            for (j = 0; j < width; j++) {
-               row[j*4+0] = val0;
-               row[j*4+1] = val1;
-               row[j*4+2] = val2;
-               row[j*4+3] = val3;
-            }
-            row += dst->stride/2;
-         }
-      }
-      break;
-   default:
-      assert(0);
-      break;
-   }
-
-   pipe->screen->surface_unmap(pipe->screen, dst);
-}
-
-
 void
 sp_init_surface_functions(struct softpipe_context *sp)
 {
-   sp->pipe.surface_copy = sp_surface_copy;
-   sp->pipe.surface_fill = sp_surface_fill;
+   sp->pipe.surface_copy = util_surface_copy;
+   sp->pipe.surface_fill = util_surface_fill;
 }
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index 01f4d0ca81..49250ec084 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -39,8 +39,9 @@
 #include "sp_tile_cache.h"
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
-#include "pipe/p_util.h"
 #include "tgsi/tgsi_exec.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 
 
 /*
@@ -50,7 +51,7 @@
  * Also note, FRAC(x) doesn't truly return the fractional part of x for x < 0.
  * Instead, if x < 0 then FRAC(x) = 1 - true_frac(x).
  */
-#define FRAC(f)  ((f) - ifloor(f))
+#define FRAC(f)  ((f) - util_ifloor(f))
 
 
 /**
@@ -99,7 +100,7 @@ nearest_texcoord(unsigned wrapMode, float s, unsigned size)
    case PIPE_TEX_WRAP_REPEAT:
       /* s limited to [0,1) */
       /* i limited to [0,size-1] */
-      i = ifloor(s * size);
+      i = util_ifloor(s * size);
       i = REMAINDER(i, size);
       return i;
    case PIPE_TEX_WRAP_CLAMP:
@@ -110,7 +111,7 @@ nearest_texcoord(unsigned wrapMode, float s, unsigned size)
       else if (s >= 1.0F)
          i = size - 1;
       else
-         i = ifloor(s * size);
+         i = util_ifloor(s * size);
       return i;
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
       {
@@ -123,7 +124,7 @@ nearest_texcoord(unsigned wrapMode, float s, unsigned size)
          else if (s > max)
             i = size - 1;
          else
-            i = ifloor(s * size);
+            i = util_ifloor(s * size);
       }
       return i;
    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
@@ -137,14 +138,14 @@ nearest_texcoord(unsigned wrapMode, float s, unsigned size)
          else if (s >= max)
             i = size;
          else
-            i = ifloor(s * size);
+            i = util_ifloor(s * size);
       }
       return i;
    case PIPE_TEX_WRAP_MIRROR_REPEAT:
       {
          const float min = 1.0F / (2.0F * size);
          const float max = 1.0F - min;
-         const int flr = ifloor(s);
+         const int flr = util_ifloor(s);
          float u;
          if (flr & 1)
             u = 1.0F - (s - (float) flr);
@@ -155,20 +156,20 @@ nearest_texcoord(unsigned wrapMode, float s, unsigned size)
          else if (u > max)
             i = size - 1;
          else
-            i = ifloor(u * size);
+            i = util_ifloor(u * size);
       }
       return i;
    case PIPE_TEX_WRAP_MIRROR_CLAMP:
       {
          /* s limited to [0,1] */
          /* i limited to [0,size-1] */
-         const float u = FABSF(s);
+         const float u = fabsf(s);
          if (u <= 0.0F)
             i = 0;
          else if (u >= 1.0F)
             i = size - 1;
          else
-            i = ifloor(u * size);
+            i = util_ifloor(u * size);
       }
       return i;
    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
@@ -177,13 +178,13 @@ nearest_texcoord(unsigned wrapMode, float s, unsigned size)
          /* i limited to [0, size-1] */
          const float min = 1.0F / (2.0F * size);
          const float max = 1.0F - min;
-         const float u = FABSF(s);
+         const float u = fabsf(s);
          if (u < min)
             i = 0;
          else if (u > max)
             i = size - 1;
          else
-            i = ifloor(u * size);
+            i = util_ifloor(u * size);
       }
       return i;
    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
@@ -192,13 +193,13 @@ nearest_texcoord(unsigned wrapMode, float s, unsigned size)
          /* i limited to [0, size-1] */
          const float min = -1.0F / (2.0F * size);
          const float max = 1.0F - min;
-         const float u = FABSF(s);
+         const float u = fabsf(s);
          if (u < min)
             i = -1;
          else if (u > max)
             i = size;
          else
-            i = ifloor(u * size);
+            i = util_ifloor(u * size);
       }
       return i;
    default:
@@ -225,7 +226,7 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
    switch (wrapMode) {
    case PIPE_TEX_WRAP_REPEAT:
       u = s * size - 0.5F;
-      *i0 = REMAINDER(ifloor(u), size);
+      *i0 = REMAINDER(util_ifloor(u), size);
       *i1 = REMAINDER(*i0 + 1, size);
       break;
    case PIPE_TEX_WRAP_CLAMP:
@@ -236,7 +237,7 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
       else
          u = s * size;
       u -= 0.5F;
-      *i0 = ifloor(u);
+      *i0 = util_ifloor(u);
       *i1 = *i0 + 1;
       break;
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
@@ -247,7 +248,7 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
       else
          u = s * size;
       u -= 0.5F;
-      *i0 = ifloor(u);
+      *i0 = util_ifloor(u);
       *i1 = *i0 + 1;
       if (*i0 < 0)
          *i0 = 0;
@@ -265,19 +266,19 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
          else
             u = s * size;
          u -= 0.5F;
-         *i0 = ifloor(u);
+         *i0 = util_ifloor(u);
          *i1 = *i0 + 1;
       }
       break;
    case PIPE_TEX_WRAP_MIRROR_REPEAT:
       {
-         const int flr = ifloor(s);
+         const int flr = util_ifloor(s);
          if (flr & 1)
             u = 1.0F - (s - (float) flr);
          else
             u = s - (float) flr;
          u = (u * size) - 0.5F;
-         *i0 = ifloor(u);
+         *i0 = util_ifloor(u);
          *i1 = *i0 + 1;
          if (*i0 < 0)
             *i0 = 0;
@@ -286,23 +287,23 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
       }
       break;
    case PIPE_TEX_WRAP_MIRROR_CLAMP:
-      u = FABSF(s);
+      u = fabsf(s);
       if (u >= 1.0F)
          u = (float) size;
       else
          u *= size;
       u -= 0.5F;
-      *i0 = ifloor(u);
+      *i0 = util_ifloor(u);
       *i1 = *i0 + 1;
       break;
    case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
-      u = FABSF(s);
+      u = fabsf(s);
       if (u >= 1.0F)
          u = (float) size;
       else
          u *= size;
       u -= 0.5F;
-      *i0 = ifloor(u);
+      *i0 = util_ifloor(u);
       *i1 = *i0 + 1;
       if (*i0 < 0)
          *i0 = 0;
@@ -313,7 +314,7 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
       {
          const float min = -1.0F / (2.0F * size);
          const float max = 1.0F - min;
-         u = FABSF(s);
+         u = fabsf(s);
          if (u <= min)
             u = min * size;
          else if (u >= max)
@@ -321,7 +322,7 @@ linear_texcoord(unsigned wrapMode, float s, unsigned size,
          else
             u *= size;
          u -= 0.5F;
-         *i0 = ifloor(u);
+         *i0 = util_ifloor(u);
          *i1 = *i0 + 1;
       }
       break;
@@ -342,12 +343,12 @@ nearest_texcoord_unnorm(unsigned wrapMode, float s, unsigned size)
    int i;
    switch (wrapMode) {
    case PIPE_TEX_WRAP_CLAMP:
-      i = ifloor(s);
+      i = util_ifloor(s);
       return CLAMP(i, 0, (int) size-1);
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
       /* fall-through */
    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      return ifloor( CLAMP(s, 0.5F, (float) size - 0.5F) );
+      return util_ifloor( CLAMP(s, 0.5F, (float) size - 0.5F) );
    default:
       assert(0);
       return 0;
@@ -367,7 +368,7 @@ linear_texcoord_unnorm(unsigned wrapMode, float s, unsigned size,
    case PIPE_TEX_WRAP_CLAMP:
       /* Not exactly what the spec says, but it matches NVIDIA output */
       s = CLAMP(s - 0.5F, 0.0f, (float) size - 1.0f);
-      *i0 = ifloor(s);
+      *i0 = util_ifloor(s);
       *i1 = *i0 + 1;
       break;
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
@@ -375,7 +376,7 @@ linear_texcoord_unnorm(unsigned wrapMode, float s, unsigned size,
    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
       s = CLAMP(s, 0.5F, (float) size - 0.5F);
       s -= 0.5F;
-      *i0 = ifloor(s);
+      *i0 = util_ifloor(s);
       *i1 = *i0 + 1;
       if (*i1 > (int) size - 1)
          *i1 = size - 1;
@@ -401,7 +402,7 @@ choose_cube_face(float rx, float ry, float rz, float *newS, float *newT)
        +rz          TEXTURE_CUBE_MAP_POSITIVE_Z_EXT    +rx    -ry   rz
        -rz          TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT    -rx    -ry   rz
    */
-   const float arx = FABSF(rx), ary = FABSF(ry), arz = FABSF(rz);
+   const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
    unsigned face;
    float sc, tc, ma;
 
@@ -476,16 +477,16 @@ compute_lambda(struct tgsi_sampler *sampler,
    {
       float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT];
       float dsdy = s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT];
-      dsdx = FABSF(dsdx);
-      dsdy = FABSF(dsdy);
+      dsdx = fabsf(dsdx);
+      dsdy = fabsf(dsdy);
       rho = MAX2(dsdx, dsdy) * sampler->texture->width[0];
    }
    if (t) {
       float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT];
       float dtdy = t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT];
       float max;
-      dtdx = FABSF(dtdx);
-      dtdy = FABSF(dtdy);
+      dtdx = fabsf(dtdx);
+      dtdy = fabsf(dtdy);
       max = MAX2(dtdx, dtdy) * sampler->texture->height[0];
       rho = MAX2(rho, max);
    }
@@ -493,13 +494,13 @@ compute_lambda(struct tgsi_sampler *sampler,
       float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT];
       float dpdy = p[QUAD_TOP_LEFT]     - p[QUAD_BOTTOM_LEFT];
       float max;
-      dpdx = FABSF(dpdx);
-      dpdy = FABSF(dpdy);
+      dpdx = fabsf(dpdx);
+      dpdy = fabsf(dpdy);
       max = MAX2(dpdx, dpdy) * sampler->texture->depth[0];
       rho = MAX2(rho, max);
    }
 
-   lambda = LOG2(rho);
+   lambda = util_fast_log2(rho);
    lambda += lodbias + sampler->state->lod_bias;
    lambda = CLAMP(lambda, sampler->state->min_lod, sampler->state->max_lod);
 
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index f775591352..cb48035771 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -33,8 +33,9 @@
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_inlines.h"
-#include "pipe/p_util.h"
 #include "pipe/p_winsys.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
 
 #include "sp_context.h"
 #include "sp_state.h"
@@ -191,7 +192,7 @@ softpipe_texture_blanket(struct pipe_screen * screen,
    spt->base.nblocksy[0] = pf_get_nblocksy(&spt->base.block, spt->base.height[0]);  
    spt->stride[0] = stride[0];
 
-   pipe_buffer_reference(screen->winsys, &spt->buffer, buffer);
+   pipe_buffer_reference(screen, &spt->buffer, buffer);
 
    return &spt->base;
 }
@@ -207,7 +208,7 @@ softpipe_texture_release(struct pipe_screen *screen,
    if (--(*pt)->refcount <= 0) {
       struct softpipe_texture *spt = softpipe_texture(*pt);
 
-      pipe_buffer_reference(screen->winsys, &spt->buffer, NULL);
+      pipe_buffer_reference(screen, &spt->buffer, NULL);
       FREE(spt);
    }
    *pt = NULL;
@@ -230,7 +231,7 @@ softpipe_get_tex_surface(struct pipe_screen *screen,
    if (ps) {
       assert(ps->refcount);
       assert(ps->winsys);
-      pipe_buffer_reference(ws, &ps->buffer, spt->buffer);
+      pipe_buffer_reference(screen, &ps->buffer, spt->buffer);
       ps->format = pt->format;
       ps->block = pt->block;
       ps->width = pt->width[level];
@@ -306,7 +307,7 @@ softpipe_surface_map( struct pipe_screen *screen,
       return NULL;
    }
 
-   map = screen->winsys->buffer_map( screen->winsys, surface->buffer, flags );
+   map = pipe_buffer_map( screen, surface->buffer, flags );
    if (map == NULL)
       return NULL;
 
@@ -330,7 +331,7 @@ static void
 softpipe_surface_unmap(struct pipe_screen *screen,
                        struct pipe_surface *surface)
 {
-   screen->winsys->buffer_unmap( screen->winsys, surface->buffer );
+   pipe_buffer_unmap( screen, surface->buffer );
 }
 
 
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c
index 57c12ffe33..b50c984513 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.c
@@ -32,9 +32,9 @@
  *    Brian Paul
  */
 
-#include "pipe/p_util.h"
 #include "pipe/p_inlines.h"
-#include "util/p_tile.h"
+#include "util/u_memory.h"
+#include "util/u_tile.h"
 #include "sp_context.h"
 #include "sp_surface.h"
 #include "sp_texture.h"