Merge commit 'origin/gallium-0.1' into gallium-0.1

Conflicts:

	src/gallium/drivers/softpipe/sp_quad_fs.c
	src/gallium/drivers/softpipe/sp_state.h
	src/gallium/drivers/softpipe/sp_state_fs.c

configs/beos

@@ -86,7 +86,7 @@ else
 endif
 
 # Directories
-SRC_DIRS = mesa glu glut/beos
+SRC_DIRS = gallium mesa glu glut/beos
 GLU_DIRS = sgi
 DRIVER_DIRS = beos
 PROGRAM_DIRS = beos samples redbook demos tests

configs/darwin

@@ -25,5 +25,5 @@ GLW_LIB_DEPS = -L/usr/X11R6/lib -lX11 -lXt $(TOP)/lib/GL.dylib
 APP_LIB_DEPS = -L$(TOP)/lib -l$(GLUT_LIB) -l$(GLU_LIB) -l$(GL_LIB) -L/usr/X11R6/lib -lX11 -lXmu -lXt -lXi -lm
 
 # omit glw lib for now:
-SRC_DIRS = mesa glu glut/glx
+SRC_DIRS = gallium mesa glu glut/glx
 

configs/darwin-x86ppc

@@ -29,5 +29,5 @@ GLW_LIB_DEPS = -L/usr/X11R6/lib -lX11 -lXt $(TOP)/lib/GL.dylib
 APP_LIB_DEPS = -L$(TOP)/lib -l$(GLUT_LIB) -l$(GLU_LIB) -l$(GL_LIB) -L/usr/X11R6/lib -lX11 -lXmu -lXt -lXi -lm
 
 # omit glw lib for now:
-SRC_DIRS = mesa glu glut/glx
+SRC_DIRS = gallium mesa glu glut/glx
 

configs/default

@@ -60,7 +60,7 @@ GLW_SOURCES = GLwDrawA.c
 
 # Directories to build
 LIB_DIR = lib
-SRC_DIRS = mesa glu glut/glx glw
+SRC_DIRS = gallium mesa glu glut/glx glw
 GLU_DIRS = sgi
 DRIVER_DIRS = x11 osmesa
 # Which subdirs under $(TOP)/progs/ to enter:

configs/freebsd-dri

@@ -36,7 +36,7 @@ GLW_LIB_DEPS = -L$(TOP)/$(LIB_DIR) -L/usr/X11R6/lib -lGL -lXt -lX11
 
 
 # Directories
-SRC_DIRS = glx/x11 mesa glu glut/glx glw
+SRC_DIRS = glx/x11 gallium mesa glu glut/glx glw
 DRIVER_DIRS = dri
 PROGRAM_DIRS = 
 WINDOW_SYSTEM=dri

configs/linux-cell

@@ -21,7 +21,7 @@ CFLAGS = $(OPT_FLAGS) -Wall -Winline -fPIC -m32 -mabi=altivec -maltivec -I. -I$(
 CXXFLAGS = $(CFLAGS)
 
 # Omitting glw here:
-SRC_DIRS = mesa glu glut/glx
+SRC_DIRS = gallium mesa glu glut/glx
 
 
 MKDEP_OPTIONS = -fdepend -Y

configs/linux-directfb

@@ -22,7 +22,7 @@ ifeq ($(HAVE_X86), yes)
 endif
 
 # Directories
-SRC_DIRS     = mesa glu glut/directfb
+SRC_DIRS     = gallium mesa glu glut/directfb
 GLU_DIRS     = sgi
 DRIVER_DIRS  = directfb
 PROGRAM_DIRS = demos directfb

configs/linux-dri

@@ -54,10 +54,10 @@ USING_EGL=0
 
 # Directories
 ifeq ($(USING_EGL), 1)
-SRC_DIRS = egl glx/x11 mesa glu glut/glx glw
+SRC_DIRS = egl glx/x11 gallium mesa glu glut/glx glw
 PROGRAM_DIRS = egl
 else
-SRC_DIRS = glx/x11 mesa glu glut/glx glw
+SRC_DIRS = glx/x11 gallium mesa glu glut/glx glw
 PROGRAM_DIRS =
 endif
 
@@ -66,4 +66,4 @@ WINDOW_SYSTEM=dri
 
 # gamma are missing because they have not been converted to use the new
 # interface.
-DRI_DIRS = intel_winsys 
+DRI_DIRS = intel 

configs/linux-dri-xcb

@@ -53,10 +53,10 @@ USING_EGL=0
 
 # Directories
 ifeq ($(USING_EGL), 1)
-SRC_DIRS = egl glx/x11 mesa glu glut/glx glw
+SRC_DIRS = egl glx/x11 gallium mesa glu glut/glx glw
 PROGRAM_DIRS = egl
 else
-SRC_DIRS = glx/x11 mesa glu glut/glx glw
+SRC_DIRS = glx/x11 gallium mesa glu glut/glx glw
 PROGRAM_DIRS =
 endif
 

configs/linux-fbdev

@@ -6,7 +6,7 @@ CONFIG_NAME = linux-fbdev
 
 CFLAGS = -O3 -ffast-math -ansi -pedantic -fPIC -D_POSIX_C_SOURCE=199309L -D_SVID_SOURCE -D_BSD_SOURCE -DPTHREADS -DUSE_GLFBDEV_DRIVER
 
-SRC_DIRS = mesa glu glut/fbdev
+SRC_DIRS = gallium mesa glu glut/fbdev
 DRIVER_DIRS = fbdev osmesa
 PROGRAM_DIRS = fbdev demos redbook samples
 

configs/linux-osmesa

@@ -14,7 +14,7 @@ CXXFLAGS = -O3 -ansi -pedantic -fPIC -ffast-math -D_POSIX_SOURCE -D_POSIX_C_SOUR
 
 
 # Directories
-SRC_DIRS = mesa glu
+SRC_DIRS = gallium mesa glu
 DRIVER_DIRS = osmesa
 PROGRAM_DIRS = osdemos
 

configs/linux-osmesa16

@@ -17,7 +17,7 @@ OSMESA_LIB_NAME = libOSMesa16.so
 
 
 # Directories
-SRC_DIRS = mesa glu
+SRC_DIRS = gallium mesa glu
 DRIVER_DIRS = osmesa
 PROGRAM_DIRS = 
 

configs/linux-osmesa16-static

@@ -18,7 +18,7 @@ OSMESA_LIB_NAME = libOSMesa16.a
 
 
 # Directories
-SRC_DIRS = mesa glu
+SRC_DIRS = gallium mesa glu
 DRIVER_DIRS = osmesa
 PROGRAM_DIRS = 
 

configs/linux-osmesa32

@@ -17,7 +17,7 @@ OSMESA_LIB_NAME = libOSMesa32.so
 
 
 # Directories
-SRC_DIRS = mesa glu
+SRC_DIRS = gallium mesa glu
 DRIVER_DIRS = osmesa
 PROGRAM_DIRS = 
 

configs/linux-solo

@@ -43,7 +43,7 @@ GLUT_LIB_DEPS = -L$(TOP)/$(LIB_DIR) -l$(GLU_LIB) -l$(GL_LIB) -lm
 APP_LIB_DEPS = -L$(TOP)/$(LIB_DIR) -l$(GLUT_LIB) -l$(GLU_LIB) -l$(GL_LIB) -lm -lpthread
 
 # Directories
-SRC_DIRS = glx/mini mesa glu glut/mini
+SRC_DIRS = glx/mini gallium mesa glu glut/mini
 DRIVER_DIRS = dri
 PROGRAM_DIRS = miniglx
 

progs/trivial/Makefile

@@ -39,6 +39,7 @@ SOURCES = \
 	linestrip-stipple-wide.c \
 	linestrip-stipple.c \
 	linestrip.c \
+	pgon-mode.c \
 	point-clip.c \
 	point-param.c \
 	point-sprite.c \

progs/trivial/pgon-mode.c

@@ -0,0 +1,136 @@
+/**
+ * Test glPolygonMode.
+ * A tri-strip w/ two tris is drawn so that the first tri is front-facing
+ * but the second tri is back-facing.
+ * Set glPolygonMode differently for the front/back faces
+ *
+ */
+
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <math.h>
+#include <GL/glut.h>
+
+static int Win;
+static GLfloat Zrot = 0;
+static GLboolean FrontFillBackUnfilled = GL_TRUE;
+static GLboolean Lines = GL_TRUE;
+
+
+static void
+Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
+
+   if (FrontFillBackUnfilled) {
+      if (Lines) {
+         printf("FrontMode = FILL, BackMode = LINE\n");
+         glPolygonMode(GL_BACK, GL_LINE);
+      }
+      else {
+         printf("FrontMode = FILL, BackMode = POINT\n");
+         glPolygonMode(GL_BACK, GL_POINT);
+      }
+      glPolygonMode(GL_FRONT, GL_FILL);
+   }
+   else {
+      if (Lines) {
+         printf("FrontMode = LINE, BackMode = FILL\n");
+         glPolygonMode(GL_FRONT, GL_LINE);
+      }
+      else {
+         printf("FrontMode = POINT, BackMode = FILL\n");
+         glPolygonMode(GL_FRONT, GL_POINT);
+      }
+      glPolygonMode(GL_BACK, GL_FILL);
+   }
+
+   glPushMatrix();
+   glRotatef(Zrot, 0, 0, 1);
+
+   glBegin(GL_TRIANGLE_STRIP);
+   glVertex2f(-1, 0);
+   glVertex2f( 1, 0);
+   glVertex2f(0,  1);
+   glVertex2f(0, -1);
+   glEnd();
+
+   glPopMatrix();
+
+   glutSwapBuffers();
+}
+
+
+static void
+Reshape(int width, int height)
+{
+   glViewport(0, 0, width, height);
+   glMatrixMode(GL_PROJECTION);
+   glLoadIdentity();
+   glFrustum(-1.0, 1.0, -1.0, 1.0, 5.0, 25.0);
+   glMatrixMode(GL_MODELVIEW);
+   glLoadIdentity();
+   glTranslatef(0.0, 0.0, -15.0);
+}
+
+
+static void
+Key(unsigned char key, int x, int y)
+{
+   const GLfloat step = 3.0;
+   (void) x;
+   (void) y;
+   switch (key) {
+   case 'p':
+      FrontFillBackUnfilled = !FrontFillBackUnfilled;
+      break;
+   case 'l':
+      Lines = !Lines;
+      break;
+   case 'z':
+      Zrot -= step;
+      break;
+   case 'Z':
+      Zrot += step;
+      break;
+   case 27:
+      glutDestroyWindow(Win);
+      exit(0);
+      break;
+   }
+   glutPostRedisplay();
+}
+
+
+static void
+Init(void)
+{
+   printf("GL_RENDERER = %s\n", (char*) glGetString(GL_RENDERER));
+
+   glLineWidth(3.0);
+   glPointSize(3.0);
+
+   glColor4f(1, 1, 1, 0.8);
+   glEnable(GL_BLEND);
+   glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
+
+   printf("Press 'p' to toggle polygon mode\n");
+}
+
+
+int
+main(int argc, char *argv[])
+{
+   glutInit(&argc, argv);
+   glutInitWindowPosition(0, 0);
+   glutInitWindowSize(400, 400);
+   glutInitDisplayMode(GLUT_RGB | GLUT_DOUBLE | GLUT_DEPTH);
+   Win = glutCreateWindow(argv[0]);
+   glutReshapeFunc(Reshape);
+   glutKeyboardFunc(Key);
+   glutDisplayFunc(Draw);
+   Init();
+   glutMainLoop();
+   return 0;
+}

src/gallium/Makefile

@@ -0,0 +1,20 @@
+TOP = ../..
+include $(TOP)/configs/current
+
+
+SUBDIRS = aux drivers
+
+
+default: subdirs
+
+
+subdirs:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE)) || exit 1 ; \
+		fi \
+	done
+
+
+clean:
+	rm -f `find . -name \*.[oa]`

src/gallium/Makefile.template

@@ -15,7 +15,10 @@ OBJECTS = $(C_SOURCES:.c=.o) \
 ### Include directories
 INCLUDES = \
 	-I. \
-	-I$(TOP)/src/mesa/pipe \
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/gallium/include/pipe \
+	-I$(TOP)/src/gallium/aux \
+	-I$(TOP)/src/gallium/drivers \
 	-I$(TOP)/src/mesa \
 	-I$(TOP)/include \
         $(DRIVER_INCLUDES)
@@ -38,7 +41,7 @@ INCLUDES = \
 default: depend symlinks $(LIBNAME)
 
 
-$(LIBNAME): $(OBJECTS) Makefile $(TOP)/src/mesa/pipe/Makefile.template
+$(LIBNAME): $(OBJECTS) Makefile $(TOP)/src/gallium/Makefile.template
 	$(TOP)/bin/mklib -o $@ -static $(OBJECTS) $(DRIVER_LIBS)
 
 

src/gallium/aux/Makefile

@@ -0,0 +1,24 @@
+TOP = ../../..
+include $(TOP)/configs/current
+
+
+ifeq ($(CONFIG_NAME), linux-llvm)
+LLVM_DIR = llvm
+endif
+
+SUBDIRS = pipebuffer $(LLVM_DIR)
+
+
+default: subdirs
+
+
+subdirs:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE)) || exit 1 ; \
+		fi \
+	done
+
+
+clean:
+	rm -f `find . -name \*.[oa]`

src/gallium/aux/draw/draw_clip.c

@@ -459,7 +459,7 @@ static void clip_reset_stipple_counter( struct draw_stage *stage )
 
 static void clip_destroy( struct draw_stage *stage )
 {
-   draw_free_tmps( stage );
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -472,7 +472,7 @@ struct draw_stage *draw_clip_stage( struct draw_context *draw )
 {
    struct clipper *clipper = CALLOC_STRUCT(clipper);
 
-   draw_alloc_tmps( &clipper->stage, MAX_CLIPPED_VERTICES+1 );
+   draw_alloc_temp_verts( &clipper->stage, MAX_CLIPPED_VERTICES+1 );
 
    clipper->stage.draw = draw;
    clipper->stage.point = clip_point;

src/gallium/aux/draw/draw_context.c

@@ -242,7 +242,7 @@ draw_convert_wide_lines(struct draw_context *draw, boolean enable)
 /**
  * Allocate space for temporary post-transform vertices, such as for clipping.
  */
-void draw_alloc_tmps( struct draw_stage *stage, unsigned nr )
+void draw_alloc_temp_verts( struct draw_stage *stage, unsigned nr )
 {
    assert(!stage->tmp);
 
@@ -260,7 +260,7 @@ void draw_alloc_tmps( struct draw_stage *stage, unsigned nr )
 }
 
 
-void draw_free_tmps( struct draw_stage *stage )
+void draw_free_temp_verts( struct draw_stage *stage )
 {
    if (stage->tmp) {
       FREE( stage->tmp[0] );

src/gallium/aux/draw/draw_cull.c

@@ -123,7 +123,7 @@ static void cull_reset_stipple_counter( struct draw_stage *stage )
 
 static void cull_destroy( struct draw_stage *stage )
 {
-   draw_free_tmps( stage );
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -135,7 +135,7 @@ struct draw_stage *draw_cull_stage( struct draw_context *draw )
 {
    struct cull_stage *cull = CALLOC_STRUCT(cull_stage);
 
-   draw_alloc_tmps( &cull->stage, 0 );
+   draw_alloc_temp_verts( &cull->stage, 0 );
 
    cull->stage.draw = draw;
    cull->stage.next = NULL;

src/gallium/aux/draw/draw_flatshade.c

@@ -176,7 +176,7 @@ static void flatshade_reset_stipple_counter( struct draw_stage *stage )
 
 static void flatshade_destroy( struct draw_stage *stage )
 {
-   draw_free_tmps( stage );
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -188,7 +188,7 @@ struct draw_stage *draw_flatshade_stage( struct draw_context *draw )
 {
    struct flat_stage *flatshade = CALLOC_STRUCT(flat_stage);
 
-   draw_alloc_tmps( &flatshade->stage, 2 );
+   draw_alloc_temp_verts( &flatshade->stage, 2 );
 
    flatshade->stage.draw = draw;
    flatshade->stage.next = NULL;

src/gallium/aux/draw/draw_offset.c

@@ -159,7 +159,7 @@ static void offset_reset_stipple_counter( struct draw_stage *stage )
 
 static void offset_destroy( struct draw_stage *stage )
 {
-   draw_free_tmps( stage );
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -171,7 +171,7 @@ struct draw_stage *draw_offset_stage( struct draw_context *draw )
 {
    struct offset_stage *offset = CALLOC_STRUCT(offset_stage);
 
-   draw_alloc_tmps( &offset->stage, 3 );
+   draw_alloc_temp_verts( &offset->stage, 3 );
 
    offset->stage.draw = draw;
    offset->stage.next = NULL;

src/gallium/aux/draw/draw_private.h

@@ -45,7 +45,7 @@
 #include "pipe/p_defines.h"
 
 #include "x86/rtasm/x86sse.h"
-#include "pipe/tgsi/exec/tgsi_exec.h"
+#include "tgsi/exec/tgsi_exec.h"
 
 
 struct gallivm_prog;
@@ -276,9 +276,9 @@ extern struct draw_stage *draw_wide_stage( struct draw_context *context );
 extern struct draw_stage *draw_validate_stage( struct draw_context *context );
 
 
-extern void draw_free_tmps( struct draw_stage *stage );
-extern void draw_reset_tmps( struct draw_stage *stage );
-extern void draw_alloc_tmps( struct draw_stage *stage, unsigned nr );
+extern void draw_free_temp_verts( struct draw_stage *stage );
+
+extern void draw_alloc_temp_verts( struct draw_stage *stage, unsigned nr );
 
 extern void draw_reset_vertex_ids( struct draw_context *draw );
 

src/gallium/aux/draw/draw_stipple.c

@@ -212,6 +212,7 @@ passthrough_tri(struct draw_stage *stage, struct prim_header *header)
 static void 
 stipple_destroy( struct draw_stage *stage )
 {
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -223,7 +224,7 @@ struct draw_stage *draw_stipple_stage( struct draw_context *draw )
 {
    struct stipple_stage *stipple = CALLOC_STRUCT(stipple_stage);
 
-   draw_alloc_tmps( &stipple->stage, 2 );
+   draw_alloc_temp_verts( &stipple->stage, 2 );
 
    stipple->stage.draw = draw;
    stipple->stage.next = NULL;

src/gallium/aux/draw/draw_twoside.c

@@ -176,7 +176,7 @@ static void twoside_reset_stipple_counter( struct draw_stage *stage )
 
 static void twoside_destroy( struct draw_stage *stage )
 {
-   draw_free_tmps( stage );
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -188,7 +188,7 @@ struct draw_stage *draw_twoside_stage( struct draw_context *draw )
 {
    struct twoside_stage *twoside = CALLOC_STRUCT(twoside_stage);
 
-   draw_alloc_tmps( &twoside->stage, 3 );
+   draw_alloc_temp_verts( &twoside->stage, 3 );
 
    twoside->stage.draw = draw;
    twoside->stage.next = NULL;

src/gallium/aux/draw/draw_unfilled.c

@@ -165,6 +165,8 @@ static void unfilled_flush( struct draw_stage *stage,
 			    unsigned flags )
 {
    stage->next->flush( stage->next, flags );
+
+   stage->tri = unfilled_first_tri;
 }
 
 
@@ -176,7 +178,7 @@ static void unfilled_reset_stipple_counter( struct draw_stage *stage )
 
 static void unfilled_destroy( struct draw_stage *stage )
 {
-   draw_free_tmps( stage );
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -188,7 +190,7 @@ struct draw_stage *draw_unfilled_stage( struct draw_context *draw )
 {
    struct unfilled_stage *unfilled = CALLOC_STRUCT(unfilled_stage);
 
-   draw_alloc_tmps( &unfilled->stage, 0 );
+   draw_alloc_temp_verts( &unfilled->stage, 0 );
 
    unfilled->stage.draw = draw;
    unfilled->stage.next = NULL;

src/gallium/aux/draw/draw_vbuf.c

@@ -523,6 +523,9 @@ static void vbuf_destroy( struct draw_stage *stage )
    if(vbuf->vf)
       draw_vf_destroy( vbuf->vf );
 
+   if (vbuf->render)
+      vbuf->render->destroy( vbuf->render );
+
    FREE( stage );
 }
 

src/gallium/aux/draw/draw_vertex.c

@@ -34,8 +34,8 @@
  */
 
 
-#include "pipe/draw/draw_private.h"
-#include "pipe/draw/draw_vertex.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vertex.h"
 
 
 /**

src/gallium/aux/draw/draw_vertex.h

@@ -25,7 +25,12 @@
  * 
  **************************************************************************/
 
-/* Author:
+/**
+ * Post-transform vertex format info.  The vertex_info struct is used by
+ * the draw_vbuf code to emit hardware-specific vertex layouts into hw
+ * vertex buffers.
+ *
+ * Author:
  *    Brian Paul
  */
 

src/gallium/aux/draw/draw_vertex_shader.c

@@ -34,13 +34,13 @@
 #include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
 #if defined(__i386__) || defined(__386__)
-#include "pipe/tgsi/exec/tgsi_sse2.h"
+#include "tgsi/exec/tgsi_sse2.h"
 #endif
 #include "draw_private.h"
 #include "draw_context.h"
 
 #include "x86/rtasm/x86sse.h"
-#include "pipe/llvm/gallivm.h"
+#include "llvm/gallivm.h"
 
 
 #define DBG_VS 0

src/gallium/aux/draw/draw_vf.c

@@ -370,7 +370,7 @@ void draw_vf_emit_vertex( struct draw_vertex_fetch *vf,
    unsigned j;
    
    for (j = 0; j < vf->attr_count; j++) {
-      if(!a[j].isconst) {
+      if (!a[j].isconst) {
 	 a[j].inputptr = (uint8_t *)&vertex->data[a[j].attrib][0];
 	 a[j].inputstride = 0; /* XXX: one-vertex-max ATM */
       }

src/gallium/aux/draw/draw_vf.h

@@ -1,5 +1,5 @@
 /*
- * Copyright 2003 Tungsten Graphics, inc.
+ * Copyright 2008 Tungsten Graphics, inc.
  * All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -20,6 +20,17 @@
  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/**
+ * Vertex fetch/store/convert code.  This functionality is used in two places:
+ * 1. Vertex fetch/convert - to grab vertex data from incoming vertex
+ *    arrays and convert to format needed by vertex shaders.
+ * 2. Vertex store/emit - to convert simple float[][4] vertex attributes
+ *    (which is the organization used throughout the draw/prim pipeline) to
+ *    hardware-specific formats and emit into hardware vertex buffers.
+ *
  *
  * Authors:
  *    Keith Whitwell <keithw@tungstengraphics.com>
@@ -33,7 +44,7 @@
 #include "pipe/p_state.h"
 
 #include "draw_vertex.h"
-#include "draw_private.h" // for vertex_header
+#include "draw_private.h" /* for vertex_header */
 
 
 enum draw_vf_attr_format {
@@ -181,6 +192,7 @@ struct draw_vf_attr_type {
    unsigned offset;
 };
 
+/** XXX this could be moved into draw_vf.c */
 struct draw_vf_fastpath {
    unsigned vertex_stride;
    unsigned attr_count;
@@ -209,6 +221,7 @@ void
 draw_vf_generate_sse_emit( struct draw_vertex_fetch *vf );
 
 
+/** XXX this type and function could probably be moved into draw_vf.c */
 struct draw_vf_format_info {
    const char *name;
    draw_vf_insert_func insert[4];

src/gallium/aux/draw/draw_wide_prims.c

@@ -175,7 +175,7 @@ static void wide_line_aa(struct draw_stage *stage,
    float *pos;
    float dx = header->v[1]->data[0][0] - header->v[0]->data[0][0];
    float dy = header->v[1]->data[0][1] - header->v[0]->data[0][1];
-   const float len = sqrt(dx * dx + dy * dy);
+   const float len = (float) sqrt(dx * dx + dy * dy);
    uint i;
 
    dx = dx * half_width / len;
@@ -400,7 +400,7 @@ static void wide_flush( struct draw_stage *stage, unsigned flags )
 }
 
 
-static void draw_reset_stipple_counter( struct draw_stage *stage )
+static void wide_reset_stipple_counter( struct draw_stage *stage )
 {
    stage->next->reset_stipple_counter( stage->next );
 }
@@ -408,7 +408,7 @@ static void draw_reset_stipple_counter( struct draw_stage *stage )
 
 static void wide_destroy( struct draw_stage *stage )
 {
-   draw_free_tmps( stage );
+   draw_free_temp_verts( stage );
    FREE( stage );
 }
 
@@ -417,7 +417,7 @@ struct draw_stage *draw_wide_stage( struct draw_context *draw )
 {
    struct wide_stage *wide = CALLOC_STRUCT(wide_stage);
 
-   draw_alloc_tmps( &wide->stage, 4 );
+   draw_alloc_temp_verts( &wide->stage, 4 );
 
    wide->stage.draw = draw;
    wide->stage.next = NULL;
@@ -425,7 +425,7 @@ struct draw_stage *draw_wide_stage( struct draw_context *draw )
    wide->stage.line = wide_first_line;
    wide->stage.tri = passthrough_tri;
    wide->stage.flush = wide_flush;
-   wide->stage.reset_stipple_counter = draw_reset_stipple_counter;
+   wide->stage.reset_stipple_counter = wide_reset_stipple_counter;
    wide->stage.destroy = wide_destroy;
 
    return &wide->stage;

src/gallium/aux/llvm/Makefile

@@ -15,7 +15,7 @@ GALLIVM_SOURCES = \
         storagesoa.cpp \
         instructionssoa.cpp
 
-INC_SOURCES = gallivm_builtins.cpp llvm_base_shader.cpp
+INC_SOURCES = gallivm_builtins.cpp
 
 CPP_SOURCES = \
 	$(GALLIVM_SOURCES)
@@ -30,7 +30,9 @@ OBJECTS = $(C_SOURCES:.c=.o) \
 ### Include directories
 INCLUDES = \
 	-I. \
-	-I$(TOP)/src/mesa/pipe \
+	-I$(TOP)/src/gallium/drivers
+	-I$(TOP)/src/gallium/aux \
+	-I$(TOP)/src/gallium/include \
 	-I$(TOP)/src/mesa \
 	-I$(TOP)/include
 
@@ -65,8 +67,6 @@ depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(INC_SOURCES)
 gallivm_builtins.cpp: llvm_builtins.c
 	clang --emit-llvm $< |llvm-as|opt -std-compile-opts|llvm2cpp -gen-contents -o=$@ -f -for=shader -funcname=createGallivmBuiltins
 
-llvm_base_shader.cpp: llvm_entry.c
-	clang --emit-llvm $< |llvm-as |opt -std-compile-opts |llvm2cpp -for=Shader -gen-module -o=$@ -f -funcname=createBaseShader
 
 # Emacs tags
 tags:
@@ -77,7 +77,7 @@ tags:
 clean:
 	-rm -f *.o */*.o *~ *.so *~ server/*.o
 	-rm -f depend depend.bak
-	-rm -f gallivm_builtins.cpp llvm_base_shader.cpp
+	-rm -f gallivm_builtins.cpp
 
 symlinks:
 

src/gallium/aux/llvm/gallivm.cpp

@@ -42,8 +42,8 @@
 #include "pipe/p_context.h"
 #include "pipe/p_shader_tokens.h"
 
-#include "pipe/tgsi/exec/tgsi_exec.h"
-#include "pipe/tgsi/util/tgsi_dump.h"
+#include "tgsi/exec/tgsi_exec.h"
+#include "tgsi/util/tgsi_dump.h"
 
 #include <llvm/Module.h>
 #include <llvm/CallingConv.h>

src/gallium/aux/llvm/gallivm_cpu.cpp

@@ -42,8 +42,8 @@
 #include "pipe/p_context.h"
 #include "pipe/p_shader_tokens.h"
 
-#include "pipe/tgsi/exec/tgsi_exec.h"
-#include "pipe/tgsi/util/tgsi_dump.h"
+#include "tgsi/exec/tgsi_exec.h"
+#include "tgsi/util/tgsi_dump.h"
 
 #include <llvm/Module.h>
 #include <llvm/CallingConv.h>

src/gallium/aux/llvm/instructionssoa.cpp

@@ -0,0 +1,121 @@
+#include "instructionssoa.h"
+
+#include "storagesoa.h"
+
+#include <llvm/Constants.h>
+
+using namespace llvm;
+
+InstructionsSoa::InstructionsSoa(llvm::Module *mod, llvm::Function *func,
+                                 llvm::BasicBlock *block, StorageSoa *storage)
+   : m_builder(block),
+     m_storage(storage),
+     m_idx(0)
+{
+}
+
+const char * InstructionsSoa::name(const char *prefix) const
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::Value * InstructionsSoa::vectorFromVals(llvm::Value *x, llvm::Value *y,
+                                              llvm::Value *z, llvm::Value *w)
+{
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   Constant *constVector = Constant::getNullValue(vectorType);
+   Value *res = m_builder.CreateInsertElement(constVector, x,
+                                              m_storage->constantInt(0),
+                                              name("vecx"));
+   res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1),
+                               name("vecxy"));
+   res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2),
+                               name("vecxyz"));
+   if (w)
+      res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3),
+                                          name("vecxyzw"));
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::arl(const std::vector<llvm::Value*> in)
+{
+   std::vector<llvm::Value*> res(4);
+
+   //Extract x's
+   llvm::Value *x1 = m_builder.CreateExtractElement(in[0],
+                                                    m_storage->constantInt(0),
+                                                    name("extractX"));
+   //cast it to an unsigned int
+   x1 = m_builder.CreateFPToUI(x1, IntegerType::get(32), name("x1IntCast"));
+
+   res[0] = x1;//vectorFromVals(x1, x2, x3, x4);
+   //only x is valid. the others shouldn't be necessary
+   /*
+   res[1] = Constant::getNullValue(m_floatVecType);
+   res[2] = Constant::getNullValue(m_floatVecType);
+   res[3] = Constant::getNullValue(m_floatVecType);
+   */
+
+   return res;
+}
+
+
+std::vector<llvm::Value*> InstructionsSoa::add(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateAdd(in1[0], in2[0], name("addx"));
+   res[1] = m_builder.CreateAdd(in1[1], in2[1], name("addy"));
+   res[2] = m_builder.CreateAdd(in1[2], in2[2], name("addz"));
+   res[3] = m_builder.CreateAdd(in1[3], in2[3], name("addw"));
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateMul(in1[0], in2[0], name("mulx"));
+   res[1] = m_builder.CreateMul(in1[1], in2[1], name("muly"));
+   res[2] = m_builder.CreateMul(in1[2], in2[2], name("mulz"));
+   res[3] = m_builder.CreateMul(in1[3], in2[3], name("mulw"));
+
+   return res;
+}
+
+void InstructionsSoa::end()
+{
+   m_builder.CreateRetVoid();
+}
+
+std::vector<llvm::Value*> InstructionsSoa::madd(const std::vector<llvm::Value*> in1,
+                                                const std::vector<llvm::Value*> in2,
+                                                const std::vector<llvm::Value*> in3)
+{
+   std::vector<llvm::Value*> res = mul(in1, in2);
+   return add(res, in3);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector)
+{
+   std::vector<llvm::Value*> res(4);
+   res[0] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(0),
+                                           name("extract1X"));
+   res[1] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(1),
+                                           name("extract2X"));
+   res[2] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(2),
+                                           name("extract3X"));
+   res[3] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(3),
+                                           name("extract4X"));
+
+   return res;
+}

src/gallium/aux/llvm/instructionssoa.h

@@ -46,17 +46,25 @@ public:
    InstructionsSoa(llvm::Module *mod, llvm::Function *func,
                    llvm::BasicBlock *block, StorageSoa *storage);
 
+   std::vector<llvm::Value*> arl(const std::vector<llvm::Value*> in);
+
    std::vector<llvm::Value*> add(const std::vector<llvm::Value*> in1,
                                  const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> madd(const std::vector<llvm::Value*> in1,
+                                  const std::vector<llvm::Value*> in2,
+                                  const std::vector<llvm::Value*> in3);
    std::vector<llvm::Value*> mul(const std::vector<llvm::Value*> in1,
                                  const std::vector<llvm::Value*> in2);
    void         end();
 
+   std::vector<llvm::Value*> extractVector(llvm::Value *vector);
 private:
    const char * name(const char *prefix) const;
+   llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
+                               llvm::Value *z, llvm::Value *w);
 private:
    llvm::LLVMFoldingBuilder  m_builder;
-
+   StorageSoa *m_storage;
 private:
    mutable int  m_idx;
    mutable char m_name[32];

src/gallium/aux/llvm/storagesoa.cpp

@@ -0,0 +1,389 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "storagesoa.h"
+
+#include "gallivm_p.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "pipe/p_debug.h"
+
+#include <llvm/BasicBlock.h>
+#include <llvm/Module.h>
+#include <llvm/Value.h>
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/InstrTypes.h>
+#include <llvm/Instructions.h>
+
+using namespace llvm;
+
+
+StorageSoa::StorageSoa(llvm::BasicBlock *block,
+                       llvm::Value *input,
+                       llvm::Value *output,
+                       llvm::Value *consts,
+                       llvm::Value *temps)
+   : m_block(block),
+     m_input(input),
+     m_output(output),
+     m_consts(consts),
+     m_temps(temps),
+     m_immediates(0),
+     m_idx(0)
+{
+}
+
+void StorageSoa::addImmediate(float *vec)
+{
+   std::vector<float> vals(4);
+   vals[0] = vec[0];
+   vals[1] = vec[1];
+   vals[2] = vec[2];
+   vals[3] = vec[3];
+   m_immediatesToFlush.push_back(vals);
+}
+
+void StorageSoa::declareImmediates()
+{
+   if (m_immediatesToFlush.empty())
+      return;
+
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   ArrayType  *vectorChannels = ArrayType::get(vectorType, 4);
+   ArrayType  *arrayType = ArrayType::get(vectorChannels, m_immediatesToFlush.size());
+
+   m_immediates = new GlobalVariable(
+      /*Type=*/arrayType,
+      /*isConstant=*/false,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Initializer=*/0, // has initializer, specified below
+      /*Name=*/name("immediates"),
+      currentModule());
+
+   std::vector<Constant*> arrayVals;
+   for (unsigned int i = 0; i < m_immediatesToFlush.size(); ++i) {
+      std::vector<float> vec = m_immediatesToFlush[i];
+      std::vector<float> vals(4);
+      std::vector<Constant*> channelArray;
+
+      vals[0] = vec[0]; vals[1] = vec[0]; vals[2] = vec[0]; vals[3] = vec[0];
+      llvm::Constant *xChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[1]; vals[1] = vec[1]; vals[2] = vec[1]; vals[3] = vec[1];
+      llvm::Constant *yChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[2]; vals[1] = vec[2]; vals[2] = vec[2]; vals[3] = vec[2];
+      llvm::Constant *zChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[3]; vals[1] = vec[3]; vals[2] = vec[3]; vals[3] = vec[3];
+      llvm::Constant *wChannel = createConstGlobalVector(vals);
+      channelArray.push_back(xChannel);
+      channelArray.push_back(yChannel);
+      channelArray.push_back(zChannel);
+      channelArray.push_back(wChannel);
+      Constant *constChannels = ConstantArray::get(vectorChannels,
+                                                   channelArray);
+      arrayVals.push_back(constChannels);
+   }
+   Constant *constArray = ConstantArray::get(arrayType, arrayVals);
+   m_immediates->setInitializer(constArray);
+
+   m_immediatesToFlush.clear();
+}
+
+llvm::Value *StorageSoa::addrElement(int idx) const
+{
+   std::map<int, llvm::Value*>::const_iterator itr = m_addresses.find(idx);
+   if (itr == m_addresses.end()) {
+      debug_printf("Trying to access invalid shader 'address'\n");
+      return 0;
+   }
+   llvm::Value * res = (*itr).second;
+
+   res = new LoadInst(res, name("addr"), false, m_block);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::inputElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_input, idx, 0);
+   res[1] = element(m_input, idx, 1);
+   res[2] = element(m_input, idx, 2);
+   res[3] = element(m_input, idx, 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::constElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+   llvm::Value *xChannel, *yChannel, *zChannel, *wChannel;
+
+   xChannel = elementPointer(m_consts, idx, 0);
+   yChannel = elementPointer(m_consts, idx, 1);
+   zChannel = elementPointer(m_consts, idx, 2);
+   wChannel = elementPointer(m_consts, idx, 3);
+
+   res[0] = alignedArrayLoad(xChannel);
+   res[1] = alignedArrayLoad(yChannel);
+   res[2] = alignedArrayLoad(zChannel);
+   res[3] = alignedArrayLoad(wChannel);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::outputElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_output, idx, 0);
+   res[1] = element(m_output, idx, 1);
+   res[2] = element(m_output, idx, 2);
+   res[3] = element(m_output, idx, 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::tempElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_temps, idx, 0);
+   res[1] = element(m_temps, idx, 1);
+   res[2] = element(m_temps, idx, 2);
+   res[3] = element(m_temps, idx, 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::immediateElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_immediates, idx, 0);
+   res[1] = element(m_immediates, idx, 1);
+   res[2] = element(m_immediates, idx, 2);
+   res[3] = element(m_immediates, idx, 3);
+
+   return res;
+}
+
+llvm::Value * StorageSoa::elementPointer(llvm::Value *ptr, llvm::Value *index,
+                                         int channel) const
+{
+   std::vector<Value*> indices;
+   if (m_immediates == ptr)
+      indices.push_back(constantInt(0));
+   indices.push_back(index);
+   indices.push_back(constantInt(channel));
+
+   GetElementPtrInst *getElem = new GetElementPtrInst(ptr,
+                                                      indices.begin(),
+                                                      indices.end(),
+                                                      name("ptr"),
+                                                      m_block);
+   return getElem;
+}
+
+llvm::Value * StorageSoa::element(llvm::Value *ptr, llvm::Value *index,
+                                  int channel) const
+{
+   llvm::Value *res = elementPointer(ptr, index, channel);
+   LoadInst *load = new LoadInst(res, name("element"), false, m_block);
+   //load->setAlignment(8);
+   return load;
+}
+
+const char * StorageSoa::name(const char *prefix) const
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::ConstantInt * StorageSoa::constantInt(int idx) const
+{
+   if (m_constInts.find(idx) != m_constInts.end()) {
+      return m_constInts[idx];
+   }
+   ConstantInt *constInt = ConstantInt::get(APInt(32,  idx));
+   m_constInts[idx] = constInt;
+   return constInt;
+}
+
+llvm::Value *StorageSoa::alignedArrayLoad(llvm::Value *val)
+{
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   PointerType *vectorPtr  = PointerType::get(vectorType, 0);
+
+   CastInst *cast = new BitCastInst(val, vectorPtr, name("toVector"), m_block);
+   LoadInst *load = new LoadInst(cast, name("alignLoad"), false, m_block);
+   load->setAlignment(8);
+   return load;
+}
+
+llvm::Module * StorageSoa::currentModule() const
+{
+    if (!m_block || !m_block->getParent())
+       return 0;
+
+    return m_block->getParent()->getParent();
+}
+
+llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &vec)
+{
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   std::vector<Constant*> immValues;
+   ConstantFP *constx = ConstantFP::get(Type::FloatTy, APFloat(vec[0]));
+   ConstantFP *consty = ConstantFP::get(Type::FloatTy, APFloat(vec[1]));
+   ConstantFP *constz = ConstantFP::get(Type::FloatTy, APFloat(vec[2]));
+   ConstantFP *constw = ConstantFP::get(Type::FloatTy, APFloat(vec[3]));
+   immValues.push_back(constx);
+   immValues.push_back(consty);
+   immValues.push_back(constz);
+   immValues.push_back(constw);
+   Constant  *constVector = ConstantVector::get(vectorType, immValues);
+
+   return constVector;
+}
+
+std::vector<llvm::Value*> StorageSoa::load(Argument type, int idx, int swizzle,
+                                           llvm::Value *indIdx)
+{
+   std::vector<llvm::Value*> val(4);
+
+   //if we have an indirect index, always use that
+   //   if not use the integer offset to create one
+   llvm::Value *realIndex = 0;
+   if (indIdx)
+      realIndex = indIdx;
+   else
+      realIndex = constantInt(idx);
+   debug_printf("XXXXXXXXX realIdx = %p, indIdx = %p\n", realIndex, indIdx);
+
+   switch(type) {
+   case Input:
+      val = inputElement(realIndex);
+      break;
+   case Output:
+      val = outputElement(realIndex);
+      break;
+   case Temp:
+      val = tempElement(realIndex);
+      break;
+   case Const:
+      val = constElement(realIndex);
+      break;
+   case Immediate:
+      val = immediateElement(realIndex);
+      break;
+   case Address:
+      debug_printf("Address not handled in the load phase!\n");
+      assert(0);
+      break;
+   }
+   if (!gallivm_is_swizzle(swizzle))
+      return val;
+
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = val[gallivm_x_swizzle(swizzle)];
+   res[1] = val[gallivm_y_swizzle(swizzle)];
+   res[2] = val[gallivm_z_swizzle(swizzle)];
+   res[3] = val[gallivm_w_swizzle(swizzle)];
+   return res;
+}
+
+void StorageSoa::store(Argument type, int idx, const std::vector<llvm::Value*> &val,
+                       int mask)
+{
+   llvm::Value *out = 0;
+   switch(type) {
+   case Output:
+      out = m_output;
+      break;
+   case Temp:
+      out = m_temps;
+      break;
+   case Input:
+      out = m_input;
+      break;
+   case Address: {
+      llvm::Value *addr = m_addresses[idx];
+      if (!addr) {
+         addAddress(idx);
+         addr = m_addresses[idx];
+         assert(addr);
+      }
+      new StoreInst(val[0], addr, false, m_block);
+      return;
+      break;
+   }
+   default:
+      debug_printf("Can't save output of this type: %d !\n", type);
+      assert(0);
+      break;
+   }
+   llvm::Value *realIndex = constantInt(idx);
+   if ((mask & TGSI_WRITEMASK_X)) {
+      llvm::Value *xChannel = elementPointer(out, realIndex, 0);
+      new StoreInst(val[0], xChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Y)) {
+      llvm::Value *yChannel = elementPointer(out, realIndex, 1);
+      new StoreInst(val[1], yChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Z)) {
+      llvm::Value *zChannel = elementPointer(out, realIndex, 2);
+      new StoreInst(val[2], zChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_W)) {
+      llvm::Value *wChannel = elementPointer(out, realIndex, 3);
+      new StoreInst(val[3], wChannel, false, m_block);
+   }
+}
+
+void StorageSoa::addAddress(int idx)
+{
+   GlobalVariable *val = new GlobalVariable(
+      /*Type=*/IntegerType::get(32),
+      /*isConstant=*/false,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Initializer=*/0, // has initializer, specified below
+      /*Name=*/name("address"),
+      currentModule());
+   val->setInitializer(Constant::getNullValue(IntegerType::get(32)));
+
+   debug_printf("adding to %d\n", idx);
+   m_addresses[idx] = val;
+}

src/gallium/aux/llvm/storagesoa.h

@@ -29,19 +29,31 @@
 #define STORAGESOA_H
 
 #include <vector>
+#include <list>
 #include <map>
 
 namespace llvm {
    class BasicBlock;
    class Constant;
    class ConstantInt;
+   class GlobalVariable;
    class LoadInst;
    class Value;
    class VectorType;
+   class Module;
 }
 
 class StorageSoa
 {
+public:
+   enum Argument {
+      Input,
+      Output,
+      Temp,
+      Const,
+      Immediate,
+      Address
+   };
 public:
    StorageSoa(llvm::BasicBlock *block,
               llvm::Value *input,
@@ -49,31 +61,35 @@ public:
               llvm::Value *consts,
               llvm::Value *temps);
 
+
+   std::vector<llvm::Value*> load(Argument type, int idx, int swizzle, 
+                                  llvm::Value *indIdx =0);
+   void store(Argument type, int idx, const std::vector<llvm::Value*> &val,
+              int mask);
+
    void addImmediate(float *vec);
+   void declareImmediates();
+
+   void addAddress(int idx);
 
    llvm::Value  * addrElement(int idx) const;
 
-   std::vector<llvm::Value*> inputElement(int idx, int swizzle, llvm::Value *indIdx =0);
-   std::vector<llvm::Value*> constElement(int idx, int swizzle, llvm::Value *indIdx =0);
-   std::vector<llvm::Value*> outputElement(int idx, int swizzle, llvm::Value *indIdx =0);
-   std::vector<llvm::Value*> tempElement(int idx, int swizzle, llvm::Value *indIdx =0);
-   std::vector<llvm::Value*> immediateElement(int idx, int swizzle);
-
-   llvm::Value *extractIndex(llvm::Value *vec);
-
-   void storeOutput(int dstIdx, const std::vector<llvm::Value*> &val,
-                         int mask);
-   void storeTemp(int idx, const std::vector<llvm::Value*> &val,
-                  int mask);
-   void storeAddress(int idx, const std::vector<llvm::Value*> &val,
-                     int mask);
+   llvm::ConstantInt *constantInt(int) const;
 private:
-   llvm::Value *elementPointer(llvm::Value *ptr, int index,
+   llvm::Value *elementPointer(llvm::Value *ptr, llvm::Value *indIdx,
                                int channel) const;
-   llvm::Value *element(llvm::Value *ptr, int index,
+   llvm::Value *element(llvm::Value *ptr, llvm::Value *idx,
                         int channel) const;
    const char *name(const char *prefix) const;
-   llvm::ConstantInt *constantInt(int) const;
+   llvm::Value  *alignedArrayLoad(llvm::Value *val);
+   llvm::Module *currentModule() const;
+   llvm::Constant  *createConstGlobalVector(const std::vector<float> &vec);
+
+   std::vector<llvm::Value*> inputElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> constElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> outputElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> tempElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> immediateElement(llvm::Value *indIdx);
 private:
    llvm::BasicBlock *m_block;
 
@@ -81,6 +97,11 @@ private:
    llvm::Value *m_output;
    llvm::Value *m_consts;
    llvm::Value *m_temps;
+   llvm::GlobalVariable *m_immediates;
+
+   std::map<int, llvm::Value*> m_addresses;
+
+   std::vector<std::vector<float> > m_immediatesToFlush;
 
    mutable std::map<int, llvm::ConstantInt*> m_constInts;
    mutable char        m_name[32];

src/gallium/aux/llvm/tgsitollvm.cpp

@@ -10,11 +10,11 @@
 
 #include "pipe/p_shader_tokens.h"
 
-#include "pipe/tgsi/util/tgsi_parse.h"
-#include "pipe/tgsi/exec/tgsi_exec.h"
-#include "pipe/tgsi/util/tgsi_util.h"
-#include "pipe/tgsi/util/tgsi_build.h"
-#include "pipe/tgsi/util/tgsi_dump.h"
+#include "tgsi/util/tgsi_parse.h"
+#include "tgsi/exec/tgsi_exec.h"
+#include "tgsi/util/tgsi_util.h"
+#include "tgsi/util/tgsi_build.h"
+#include "tgsi/util/tgsi_dump.h"
 
 
 #include <llvm/Module.h>
@@ -44,7 +44,6 @@
 #include <iostream>
 
 using namespace llvm;
-#include "llvm_base_shader.cpp"
 
 static inline FunctionType *vertexShaderFunctionType()
 {
@@ -149,10 +148,14 @@ translate_declaration(struct gallivm_ir *prog,
 static void
 translate_declarationir(struct gallivm_ir *,
                       llvm::Module *,
-                      StorageSoa *,
-                      struct tgsi_full_declaration *,
+                      StorageSoa *storage,
+                      struct tgsi_full_declaration *decl,
                       struct tgsi_full_declaration *)
 {
+   if (decl->Declaration.File == TGSI_FILE_ADDRESS) {
+      int idx = decl->u.DeclarationRange.First;
+      storage->addAddress(idx);
+   }
 }
 
 static void
@@ -704,18 +707,22 @@ translate_instructionir(llvm::Module *module,
 
       if (src->SrcRegister.Indirect) {
          indIdx = storage->addrElement(src->SrcRegisterInd.Index);
-         indIdx = storage->extractIndex(indIdx);
       }
       if (src->SrcRegister.File == TGSI_FILE_CONSTANT) {
-         val = storage->constElement(src->SrcRegister.Index, swizzle, indIdx);
+         val = storage->load(StorageSoa::Const,
+                             src->SrcRegister.Index, swizzle, indIdx);
       } else if (src->SrcRegister.File == TGSI_FILE_INPUT) {
-         val = storage->inputElement(src->SrcRegister.Index, swizzle, indIdx);
+         val = storage->load(StorageSoa::Input,
+                             src->SrcRegister.Index, swizzle, indIdx);
       } else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) {
-         val = storage->tempElement(src->SrcRegister.Index, swizzle);
+         val = storage->load(StorageSoa::Temp,
+                             src->SrcRegister.Index, swizzle, indIdx);
       } else if (src->SrcRegister.File == TGSI_FILE_OUTPUT) {
-         val = storage->outputElement(src->SrcRegister.Index, swizzle, indIdx);
+         val = storage->load(StorageSoa::Output,
+                             src->SrcRegister.Index, swizzle, indIdx);
       } else if (src->SrcRegister.File == TGSI_FILE_IMMEDIATE) {
-         val = storage->immediateElement(src->SrcRegister.Index, swizzle);
+         val = storage->load(StorageSoa::Immediate,
+                             src->SrcRegister.Index, swizzle, indIdx);
       } else {
          fprintf(stderr, "ERROR: not supported llvm source %d\n", src->SrcRegister.File);
          return;
@@ -727,6 +734,7 @@ translate_instructionir(llvm::Module *module,
    std::vector<llvm::Value*> out(4);
    switch (inst->Instruction.Opcode) {
    case TGSI_OPCODE_ARL: {
+      out = instr->arl(inputs[0]);
    }
       break;
    case TGSI_OPCODE_MOV: {
@@ -776,6 +784,7 @@ translate_instructionir(llvm::Module *module,
    }
       break;
    case TGSI_OPCODE_MAD: {
+      out = instr->madd(inputs[0], inputs[1], inputs[2]);
    }
       break;
    case TGSI_OPCODE_SUB: {
@@ -1059,11 +1068,14 @@ translate_instructionir(llvm::Module *module,
       struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
 
       if (dst->DstRegister.File == TGSI_FILE_OUTPUT) {
-         storage->storeOutput(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+         storage->store(StorageSoa::Output,
+                        dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
       } else if (dst->DstRegister.File == TGSI_FILE_TEMPORARY) {
-         storage->storeTemp(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+         storage->store(StorageSoa::Temp,
+                        dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
       } else if (dst->DstRegister.File == TGSI_FILE_ADDRESS) {
-         storage->storeAddress(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+         storage->store(StorageSoa::Address,
+                        dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
       } else {
          fprintf(stderr, "ERROR: unsupported LLVM destination!");
          assert(!"wrong destination");
@@ -1074,7 +1086,7 @@ translate_instructionir(llvm::Module *module,
 llvm::Module *
 tgsi_to_llvm(struct gallivm_ir *ir, const struct tgsi_token *tokens)
 {
-   llvm::Module *mod = createBaseShader();
+   llvm::Module *mod = new Module("shader");
    struct tgsi_parse_context parse;
    struct tgsi_full_instruction fi;
    struct tgsi_full_declaration fd;
@@ -1191,6 +1203,7 @@ llvm::Module * tgsi_to_llvmir(struct gallivm_ir *ir,
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
+         storage.declareImmediates();
          translate_instructionir(mod, &storage, &instr,
                                  &parse.FullToken.FullInstruction,
                                  &fi, instno);

src/gallium/aux/pipebuffer/Makefile

@@ -17,7 +17,7 @@ C_SOURCES = \
 
 ASM_SOURCES = 
 
-include ../Makefile.template
+include ../../Makefile.template
 
 symlinks:
 

src/gallium/aux/pipebuffer/pb_buffer_malloc.c

@@ -110,7 +110,7 @@ pb_malloc_buffer_create(size_t size,
    buf = CALLOC_STRUCT(malloc_buffer);
    if(!buf)
       return NULL;
-   
+
    buf->base.base.refcount = 1;
    buf->base.base.alignment = desc->alignment;
    buf->base.base.usage = desc->usage;
@@ -119,7 +119,7 @@ pb_malloc_buffer_create(size_t size,
 
    buf->data = align_malloc(size, desc->alignment < sizeof(void*) ? sizeof(void*) : desc->alignment);
    if(!buf->data) {
-      FREE(buf);
+      align_free(buf);
       return NULL;
    }
 

src/gallium/aux/tgsi/exec/tgsi_exec.c

@@ -54,8 +54,8 @@
 #include "pipe/p_state.h"
 #include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
-#include "pipe/tgsi/util/tgsi_parse.h"
-#include "pipe/tgsi/util/tgsi_util.h"
+#include "tgsi/util/tgsi_parse.h"
+#include "tgsi/util/tgsi_util.h"
 #include "tgsi_exec.h"
 
 #define TILE_TOP_LEFT     0
@@ -176,13 +176,13 @@ tgsi_exec_machine_bind_shader(
          if (numDeclarations == maxDeclarations) {
             declarations = REALLOC(declarations,
                                    maxDeclarations
-                                   * sizeof(struct tgsi_full_instruction),
+                                   * sizeof(struct tgsi_full_declaration),
                                    (maxDeclarations + 10)
-                                   * sizeof(struct tgsi_full_instruction));
+                                   * sizeof(struct tgsi_full_declaration));
             maxDeclarations += 10;
          }
          memcpy(declarations + numDeclarations,
-                &parse.FullToken.FullInstruction,
+                &parse.FullToken.FullDeclaration,
                 sizeof(declarations[0]));
          numDeclarations++;
          break;
@@ -1337,9 +1337,12 @@ exec_tex(struct tgsi_exec_machine *mach,
 }
 
 
-
+/**
+ * Evaluate a constant-valued coefficient at the position of the
+ * current quad.
+ */
 static void
-constant_interpolation(
+eval_constant_coef(
    struct tgsi_exec_machine *mach,
    unsigned attrib,
    unsigned chan )
@@ -1351,8 +1354,12 @@ constant_interpolation(
    }
 }
 
+/**
+ * Evaluate a linear-valued coefficient at the position of the
+ * current quad.
+ */
 static void
-linear_interpolation(
+eval_linear_coef(
    struct tgsi_exec_machine *mach,
    unsigned attrib,
    unsigned chan )
@@ -1368,8 +1375,12 @@ linear_interpolation(
    mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady;
 }
 
+/**
+ * Evaluate a perspective-valued coefficient at the position of the
+ * current quad.
+ */
 static void
-perspective_interpolation(
+eval_perspective_coef(
    struct tgsi_exec_machine *mach,
    unsigned attrib,
    unsigned chan )
@@ -1388,7 +1399,7 @@ perspective_interpolation(
 }
 
 
-typedef void (* interpolation_func)(
+typedef void (* eval_coef_func)(
    struct tgsi_exec_machine *mach,
    unsigned attrib,
    unsigned chan );
@@ -1401,7 +1412,7 @@ exec_declaration(
    if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
       if( decl->Declaration.File == TGSI_FILE_INPUT ) {
          unsigned first, last, mask;
-         interpolation_func interp;
+         eval_coef_func eval;
 
          assert( decl->Declaration.Declare == TGSI_DECLARE_RANGE );
 
@@ -1411,15 +1422,15 @@ exec_declaration(
 
          switch( decl->Interpolation.Interpolate ) {
          case TGSI_INTERPOLATE_CONSTANT:
-            interp = constant_interpolation;
+            eval = eval_constant_coef;
             break;
 
          case TGSI_INTERPOLATE_LINEAR:
-            interp = linear_interpolation;
+            eval = eval_linear_coef;
             break;
 
          case TGSI_INTERPOLATE_PERSPECTIVE:
-            interp = perspective_interpolation;
+            eval = eval_perspective_coef;
             break;
 
          default:
@@ -1431,7 +1442,7 @@ exec_declaration(
 
             for( i = first; i <= last; i++ ) {
                for( j = 0; j < NUM_CHANNELS; j++ ) {
-                  interp( mach, i, j );
+                  eval( mach, i, j );
                }
             }
          }
@@ -1441,7 +1452,7 @@ exec_declaration(
             for( j = 0; j < NUM_CHANNELS; j++ ) {
                if( mask & (1 << j) ) {
                   for( i = first; i <= last; i++ ) {
-                     interp( mach, i, j );
+                     eval( mach, i, j );
                   }
                }
             }

src/gallium/aux/tgsi/exec/tgsi_sse2.c

@@ -27,8 +27,8 @@
 
 #include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
-#include "pipe/tgsi/util/tgsi_parse.h"
-#include "pipe/tgsi/util/tgsi_util.h"
+#include "tgsi/util/tgsi_parse.h"
+#include "tgsi/util/tgsi_util.h"
 #include "tgsi_exec.h"
 #include "tgsi_sse2.h"
 

src/gallium/aux/tgsi/util/tgsi_transform.c

@@ -0,0 +1,199 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI program transformation utility.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "tgsi_transform.h"
+
+
+
+static void
+emit_instruction(struct tgsi_transform_context *ctx,
+                 const struct tgsi_full_instruction *inst)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_instruction(inst,
+                                     ctx->tokens_out + ti,
+                                     ctx->header,
+                                     ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+static void
+emit_declaration(struct tgsi_transform_context *ctx,
+                 const struct tgsi_full_declaration *decl)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_declaration(decl,
+                                     ctx->tokens_out + ti,
+                                     ctx->header,
+                                     ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+static void
+emit_immediate(struct tgsi_transform_context *ctx,
+               const struct tgsi_full_immediate *imm)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_immediate(imm,
+                                   ctx->tokens_out + ti,
+                                   ctx->header,
+                                   ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+
+/**
+ * Apply user-defined transformations to the input shader to produce
+ * the output shader.
+ * For example, a register search-and-replace operation could be applied
+ * by defining a transform_instruction() callback that examined and changed
+ * the instruction src/dest regs.
+ *
+ * \return number of tokens emitted
+ */
+int
+tgsi_transform_shader(const struct tgsi_token *tokens_in,
+                      struct tgsi_token *tokens_out,
+                      uint max_tokens_out,
+                      struct tgsi_transform_context *ctx)
+{
+   uint procType;
+
+   /* input shader */
+   struct tgsi_parse_context parse;
+
+   /* output shader */
+   struct tgsi_processor *processor;
+
+
+   /**
+    ** callback context init
+    **/
+   ctx->emit_instruction = emit_instruction;
+   ctx->emit_declaration = emit_declaration;
+   ctx->emit_immediate = emit_immediate;
+   ctx->tokens_out = tokens_out;
+   ctx->max_tokens_out = max_tokens_out;
+
+
+   /**
+    ** Setup to begin parsing input shader
+    **/
+   if (tgsi_parse_init( &parse, tokens_in ) != TGSI_PARSE_OK) {
+      debug_printf("tgsi_parse_init() failed in tgsi_transform_shader()!\n");
+      return -1;
+   }
+   procType = parse.FullHeader.Processor.Processor;
+   assert(procType == TGSI_PROCESSOR_FRAGMENT ||
+          procType == TGSI_PROCESSOR_VERTEX ||
+          procType == TGSI_PROCESSOR_GEOMETRY);
+
+
+   /**
+    **  Setup output shader
+    **/
+   *(struct tgsi_version *) &tokens_out[0] = tgsi_build_version();
+
+   ctx->header = (struct tgsi_header *) (tokens_out + 1);
+   *ctx->header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) (tokens_out + 2);
+   *processor = tgsi_build_processor( procType, ctx->header );
+
+   ctx->ti = 3;
+
+
+   /**
+    ** Loop over incoming program tokens/instructions
+    */
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         {
+            struct tgsi_full_instruction *fullinst
+               = &parse.FullToken.FullInstruction;
+
+            if (ctx->transform_instruction)
+               ctx->transform_instruction(ctx, fullinst);
+            else
+               ctx->emit_instruction(ctx, fullinst);
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         {
+            struct tgsi_full_declaration *fulldecl
+               = &parse.FullToken.FullDeclaration;
+
+            if (ctx->transform_declaration)
+               ctx->transform_declaration(ctx, fulldecl);
+            else
+               ctx->emit_declaration(ctx, fulldecl);
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         {
+            struct tgsi_full_immediate *fullimm
+               = &parse.FullToken.FullImmediate;
+
+            if (ctx->transform_immediate)
+               ctx->transform_immediate(ctx, fullimm);
+            else
+               ctx->emit_immediate(ctx, fullimm);
+         }
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+
+   if (ctx->epilog) {
+      ctx->epilog(ctx);
+   }
+
+   tgsi_parse_free (&parse);
+
+   return ctx->ti;
+}

src/gallium/aux/tgsi/util/tgsi_transform.h

@@ -0,0 +1,93 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_TRANSFORM_H
+#define TGSI_TRANSFORM_H
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/util/tgsi_parse.h"
+#include "tgsi/util/tgsi_build.h"
+
+
+
+/**
+ * Subclass this to add caller-specific data
+ */
+struct tgsi_transform_context
+{
+/**** PUBLIC ***/
+
+   /**
+    * User-defined callbacks invoked per instruction.
+    */
+   void (*transform_instruction)(struct tgsi_transform_context *ctx,
+                                 struct tgsi_full_instruction *inst);
+
+   void (*transform_declaration)(struct tgsi_transform_context *ctx,
+                                 struct tgsi_full_declaration *decl);
+
+   void (*transform_immediate)(struct tgsi_transform_context *ctx,
+                               struct tgsi_full_immediate *imm);
+
+   /**
+    * Called at end of input program to allow caller to append extra
+    * instructions.  Return number of tokens emitted.
+    */
+   void (*epilog)(struct tgsi_transform_context *ctx);
+
+
+/*** PRIVATE ***/
+
+   /**
+    * These are setup by tgsi_transform_shader() and cannot be overridden.
+    * Meant to be called from in the above user callback functions.
+    */
+   void (*emit_instruction)(struct tgsi_transform_context *ctx,
+                            const struct tgsi_full_instruction *inst);
+   void (*emit_declaration)(struct tgsi_transform_context *ctx,
+                            const struct tgsi_full_declaration *decl);
+   void (*emit_immediate)(struct tgsi_transform_context *ctx,
+                          const struct tgsi_full_immediate *imm);
+
+   struct tgsi_header *header;
+   uint max_tokens_out;
+   struct tgsi_token *tokens_out;
+   uint ti;
+};
+
+
+
+extern int
+tgsi_transform_shader(const struct tgsi_token *tokens_in,
+                      struct tgsi_token *tokens_out,
+                      uint max_tokens_out,
+                      struct tgsi_transform_context *ctx);
+
+
+#endif /* TGSI_TRANSFORM_H */