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

29 files changed, 512 insertions, 314 deletions

diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index 3b687bb8689..1433a4925fa 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -134,6 +134,8 @@ extern void
 cell_vertex_shader_queue_flush(struct draw_context *draw);
 
 
+/* XXX find a better home for this */
+extern void cell_update_vertex_fetch(struct draw_context *draw);
 
 
 #endif /* CELL_CONTEXT_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index cbd387f0142..c839fb4d12d 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -38,6 +38,7 @@
 #include "cell_context.h"
 #include "cell_draw_arrays.h"
 #include "cell_state.h"
+#include "cell_flush.h"
 
 #include "draw/draw_context.h"
 
diff --git a/src/gallium/drivers/cell/ppu/cell_flush.h b/src/gallium/drivers/cell/ppu/cell_flush.h
index eda351b1cbc..7f940ae76b6 100644
--- a/src/gallium/drivers/cell/ppu/cell_flush.h
+++ b/src/gallium/drivers/cell/ppu/cell_flush.h
@@ -35,4 +35,8 @@ cell_flush(struct pipe_context *pipe, unsigned flags);
 extern void
 cell_flush_int(struct pipe_context *pipe, unsigned flags);
 
+extern void
+cell_flush_buffer_range(struct cell_context *cell, void *ptr,
+			unsigned size);
+
 #endif
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
index f7ef72e5a2c..f5c27852c14 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
@@ -55,7 +55,6 @@ cell_vertex_shader_queue_flush(struct draw_context *draw)
    struct cell_command_vs *const vs = &cell_global.command[0].vs;
    uint64_t *batch;
    struct cell_array_info *array_info;
-   struct cell_shader_info *shader_info;
    unsigned i, j;
    struct cell_attribute_fetch_code *cf;
 
@@ -123,12 +122,12 @@ cell_vertex_shader_queue_flush(struct draw_context *draw)
 
       for (j = 0; j < n; j++) {
          vs->elts[j] = draw->vs.queue[i + j].elt;
-         vs->vOut[j] = (uintptr_t) draw->vs.queue[i + j].dest;
+         vs->vOut[j] = (uintptr_t) draw->vs.queue[i + j].vertex;
       }
 
       for (/* empty */; j < SPU_VERTS_PER_BATCH; j++) {
          vs->elts[j] = vs->elts[0];
-         vs->vOut[j] = vs->vOut[0];
+         vs->vOut[j] = (uintptr_t) draw->vs.queue[i + j].vertex;
       }
 
       vs->num_elts = n;
@@ -137,5 +136,6 @@ cell_vertex_shader_queue_flush(struct draw_context *draw)
       cell_flush_int(& cell->pipe, PIPE_FLUSH_WAIT);
    }
 
+   draw->vs.post_nr = draw->vs.queue_nr;
    draw->vs.queue_nr = 0;
 }
diff --git a/src/gallium/drivers/cell/spu/spu_dcache.c b/src/gallium/drivers/cell/spu/spu_dcache.c
index 698a5790bb0..a1701d80d18 100644
--- a/src/gallium/drivers/cell/spu/spu_dcache.c
+++ b/src/gallium/drivers/cell/spu/spu_dcache.c
@@ -33,7 +33,7 @@
 #define CACHE_NAME            data
 #define CACHED_TYPE           qword
 #define CACHE_TYPE            CACHE_TYPE_RO
-#define CACHE_SET_TAGID(set)  TAG_VERTEX_BUFFER
+#define CACHE_SET_TAGID(set)  (((set) & 0x03) + TAG_DCACHE0)
 #define CACHE_LOG2NNWAY       2
 #define CACHE_LOG2NSETS       6
 #include <cache-api.h>
@@ -49,43 +49,57 @@
 
 /**
  * Fetch between arbitrary number of bytes from an unaligned address
+ *
+ * \param dst   Destination data buffer
+ * \param ea    Main memory effective address of source data
+ * \param size  Number of bytes to read
+ *
+ * \warning
+ * As is hinted by the type of the \c dst pointer, this function writes
+ * multiples of 16-bytes.
  */
 void
 spu_dcache_fetch_unaligned(qword *dst, unsigned ea, unsigned size)
 {
    const int shift = ea & 0x0f;
-   const unsigned aligned_start_ea = ea & ~0x0f;
-   const unsigned aligned_end_ea = ROUNDUP16(ea + size);
-   const unsigned num_entries = (aligned_end_ea - aligned_start_ea) / 16;
+   const unsigned read_size = ROUNDUP16(size + shift);
+   const unsigned last_read = ROUNDUP16(ea + size);
+   const qword *const last_write = dst + (ROUNDUP16(size) / 16);
    unsigned i;
 
 
    if (shift == 0) {
       /* Data is already aligned.  Fetch directly into the destination buffer.
        */
-      for (i = 0; i < num_entries; i++) {
-         dst[i] = cache_rd(data, ea + (i * 16));
+      for (i = 0; i < size; i += 16) {
+         *(dst++) = cache_rd(data, ea + i);
       }
    } else {
-      qword tmp[2] ALIGN16_ATTRIB;
-
+      qword hi;
 
-      tmp[0] = cache_rd(data, (ea & ~0x0f));
-      for (i = 0; i < (num_entries & ~1); i++) {
-         const unsigned curr = i & 1;
-         const unsigned next = curr ^ 1;
 
-         tmp[next] = cache_rd(data, (ea & ~0x0f) + (next * 16));
-
-         dst[i] = si_or((qword) spu_slqwbyte(tmp[curr], shift),
-                        (qword) spu_rlmaskqwbyte(tmp[next], shift - 16));
+      /* Please exercise extreme caution when modifying this code.  This code
+       * must not read past the end of the page containing the source data,
+       * and it must not write more than ((size + 15) / 16) qwords to the
+       * destination buffer.
+       */
+      ea &= ~0x0f;
+      hi = cache_rd(data, ea);
+      for (i = 16; i < read_size; i += 16) {
+         qword lo = cache_rd(data, ea + i);
+
+         *(dst++) = si_or((qword) spu_slqwbyte(hi, shift),
+                          (qword) spu_rlmaskqwbyte(lo, shift - 16));
+         hi = lo;
       }
 
-      if (i < num_entries) {
-         dst[i] = si_or((qword) spu_slqwbyte(tmp[(i & 1)], shift),
-                        si_il(0));
+      if (dst != last_write) {
+         *(dst++) = si_or((qword) spu_slqwbyte(hi, shift), si_il(0));
       }
    }
+   
+   ASSERT((ea + i) == last_read);
+   ASSERT(dst == last_write);
 }
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_exec.c b/src/gallium/drivers/cell/spu/spu_exec.c
index cf81bee8fde..1560c0f1574 100644
--- a/src/gallium/drivers/cell/spu/spu_exec.c
+++ b/src/gallium/drivers/cell/spu/spu_exec.c
@@ -50,8 +50,6 @@
  *   Brian Paul
  */
 
-#include <libmisc.h>
-#include <spu_mfcio.h>
 #include <transpose_matrix4x4.h>
 #include <simdmath/ceilf4.h>
 #include <simdmath/cosf4.h>
@@ -151,6 +149,7 @@ spu_exec_machine_init(struct spu_exec_machine *mach,
    const qword zero = si_il(0);
    const qword not_zero = si_il(~0);
 
+   (void) numSamplers;
    mach->Samplers = samplers;
    mach->Processor = processor;
    mach->Addrs = &mach->Temps[TGSI_EXEC_NUM_TEMPS];
@@ -659,9 +658,10 @@ fetch_texel( struct spu_sampler *sampler,
    qword rgba[4];
    qword out[4];
 
-   sampler->get_samples(sampler, s->f, t->f, p->f, lodbias, (float *) rgba);
+   sampler->get_samples(sampler, s->f, t->f, p->f, lodbias, 
+			(float (*)[4]) rgba);
 
-   _transpose_matrix4x4(out, rgba);
+   _transpose_matrix4x4((vec_float4 *) out, (vec_float4 *) rgba);
    r->q = out[0];
    g->q = out[1];
    b->q = out[2];
diff --git a/src/gallium/drivers/cell/spu/spu_main.c b/src/gallium/drivers/cell/spu/spu_main.c
index 1136dba62d5..cc4bafdb3ac 100644
--- a/src/gallium/drivers/cell/spu/spu_main.c
+++ b/src/gallium/drivers/cell/spu/spu_main.c
@@ -38,6 +38,7 @@
 #include "spu_tile.h"
 //#include "spu_test.h"
 #include "spu_vertex_shader.h"
+#include "spu_dcache.h"
 #include "cell/common.h"
 #include "pipe/p_defines.h"
 
@@ -434,7 +435,7 @@ cmd_batch(uint opcode)
          pos += (1 + ROUNDUP8(sizeof(struct pipe_viewport_state)) / 8);
          break;
       case CELL_CMD_STATE_UNIFORMS:
-         draw.constants = (float (*)[4]) (uintptr_t) buffer[pos + 1];
+         draw.constants = (const float (*)[4]) (uintptr_t) buffer[pos + 1];
          pos += 2;
          break;
       case CELL_CMD_STATE_VS_ARRAY_INFO:
@@ -583,7 +584,7 @@ main(main_param_t speid, main_param_t argp)
    one_time_init();
 
    if (Debug)
-      printf("SPU: main() speid=%lu\n", speid);
+      printf("SPU: main() speid=%lu\n", (unsigned long) speid);
 
    mfc_get(&spu.init,  /* dest */
            (unsigned int) argp, /* src */
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index 5c95d112ac1..d14f1abbe74 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -131,7 +131,10 @@ extern boolean Debug;
 #define TAG_BATCH_BUFFER      17
 #define TAG_MISC              18
 #define TAG_TEXTURE_TILE      19
-#define TAG_INSTRUCTION_FETCH 20
+#define TAG_DCACHE0           20
+#define TAG_DCACHE1           21
+#define TAG_DCACHE2           22
+#define TAG_DCACHE3           23
 
 
 
diff --git a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
index f7e4e653e31..219fd90cc0e 100644
--- a/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
+++ b/src/gallium/drivers/cell/spu/spu_vertex_fetch.c
@@ -32,8 +32,6 @@
   *   Ian Romanick <[email protected]>
   */
 
-#include <spu_mfcio.h>
-
 #include "pipe/p_util.h"
 #include "pipe/p_state.h"
 #include "pipe/p_shader_tokens.h"
diff --git a/src/gallium/drivers/i915simple/i915_context.c b/src/gallium/drivers/i915simple/i915_context.c
index acfa3494397..c3955bbd2dd 100644
--- a/src/gallium/drivers/i915simple/i915_context.c
+++ b/src/gallium/drivers/i915simple/i915_context.c
@@ -298,10 +298,12 @@ struct pipe_context *i915_create( struct pipe_winsys *pipe_winsys,
    i915_init_string_functions(i915);
    i915_init_texture_functions(i915);
 
+   draw_install_aaline_stage(i915->draw, &i915->pipe);
+   draw_install_aapoint_stage(i915->draw, &i915->pipe);
+
    i915->pci_id = pci_id;
    i915->flags.is_i945 = is_i945;
 
-
    i915->dirty = ~0;
    i915->hardware_dirty = ~0;
 
diff --git a/src/gallium/drivers/i915simple/i915_context.h b/src/gallium/drivers/i915simple/i915_context.h
index 2d876925b2c..d32dded6bdc 100644
--- a/src/gallium/drivers/i915simple/i915_context.h
+++ b/src/gallium/drivers/i915simple/i915_context.h
@@ -79,6 +79,40 @@
 #define I915_MAX_CONSTANT  32
 
 
+/** See constant_flags[] below */
+#define I915_CONSTFLAG_USER 0x1f
+
+
+/**
+ * Subclass of pipe_shader_state
+ */
+struct i915_fragment_shader
+{
+   struct pipe_shader_state state;
+   uint *program;
+   uint program_len;
+
+   /**
+    * constants introduced during translation.
+    * These are placed at the end of the constant buffer and grow toward
+    * the beginning (eg: slot 31, 30 29, ...)
+    * User-provided constants start at 0.
+    * This allows both types of constants to co-exist (until there's too many)
+    * and doesn't require regenerating/changing the fragment program to
+    * shuffle constants around.
+    */
+   uint num_constants;
+   float constants[I915_MAX_CONSTANT][4];
+
+   /**
+    * Status of each constant
+    * if I915_CONSTFLAG_PARAM, the value must be taken from the corresponding
+    * slot of the user's constant buffer. (set by pipe->set_constant_buffer())
+    * Else, the bitmask indicates which components are occupied by immediates.
+    */
+   ubyte constant_flags[I915_MAX_CONSTANT];
+};
+
 
 struct i915_cache_context;
 
@@ -93,11 +127,6 @@ struct i915_state
    float constants[PIPE_SHADER_TYPES][I915_MAX_CONSTANT][4];
    /** number of constants passed in through a constant buffer */
    uint num_user_constants[PIPE_SHADER_TYPES];
-   /** user constants, plus extra constants from shader translation */
-   uint num_constants[PIPE_SHADER_TYPES];
-
-   uint *program;
-   uint program_len;
 
    /* texture sampler state */
    unsigned sampler[I915_TEX_UNITS][3];
@@ -187,7 +216,8 @@ struct i915_context
    const struct i915_sampler_state         *sampler[PIPE_MAX_SAMPLERS];
    const struct i915_depth_stencil_state   *depth_stencil;
    const struct i915_rasterizer_state      *rasterizer;
-   const struct pipe_shader_state *fs;
+
+   struct i915_fragment_shader *fs;
 
    struct pipe_blend_color blend_color;
    struct pipe_clip_state clip;
@@ -233,6 +263,7 @@ struct i915_context
 #define I915_NEW_TEXTURE       0x800
 #define I915_NEW_CONSTANTS     0x1000
 #define I915_NEW_VBO           0x2000
+#define I915_NEW_VS            0x4000
 
 
 /* Driver's internally generated state flags:
diff --git a/src/gallium/drivers/i915simple/i915_flush.c b/src/gallium/drivers/i915simple/i915_flush.c
index 3c2069b8273..96a54281f11 100644
--- a/src/gallium/drivers/i915simple/i915_flush.c
+++ b/src/gallium/drivers/i915simple/i915_flush.c
@@ -31,6 +31,7 @@
 
 
 #include "pipe/p_defines.h"
+#include "draw/draw_context.h"
 #include "i915_context.h"
 #include "i915_reg.h"
 #include "i915_batch.h"
@@ -44,6 +45,8 @@ static void i915_flush( struct pipe_context *pipe,
 {
    struct i915_context *i915 = i915_context(pipe);
 
+   draw_flush(i915->draw);
+
    /* Do we need to emit an MI_FLUSH command to flush the hardware
     * caches?
     */
diff --git a/src/gallium/drivers/i915simple/i915_fpc.h b/src/gallium/drivers/i915simple/i915_fpc.h
index 8c7b68aefb5..250dfe6dbf0 100644
--- a/src/gallium/drivers/i915simple/i915_fpc.h
+++ b/src/gallium/drivers/i915simple/i915_fpc.h
@@ -44,9 +44,16 @@
  * Program translation state
  */
 struct i915_fp_compile {
-   const struct pipe_shader_state *shader;
+   struct i915_fragment_shader *shader;  /* the shader we're compiling */
 
-   struct vertex_info *vertex_info;
+   boolean used_constants[I915_MAX_CONSTANT];
+
+   /** maps TGSI immediate index to constant slot */
+   uint num_immediates;
+   uint immediates_map[I915_MAX_CONSTANT];
+   float immediates[I915_MAX_CONSTANT][4];
+
+   boolean first_instruction;
 
    uint declarations[I915_PROGRAM_SIZE];
    uint program[I915_PROGRAM_SIZE];
@@ -57,11 +64,6 @@ struct i915_fp_compile {
    uint output_semantic_name[PIPE_MAX_SHADER_OUTPUTS];
    uint output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
 
-   /** points into the i915->current.constants array: */
-   float (*constants)[4];
-   uint num_constants;
-   uint constant_flags[I915_MAX_CONSTANT]; /**< status of each constant */
-
    uint *csr;            /**< Cursor, points into program. */
 
    uint *decl;           /**< Cursor, points into declarations. */
@@ -155,7 +157,9 @@ swizzle(int reg, uint x, uint y, uint z, uint w)
 /***********************************************************************
  * Public interface for the compiler
  */
-extern void i915_translate_fragment_program( struct i915_context *i915 );
+extern void
+i915_translate_fragment_program( struct i915_context *i915,
+                                 struct i915_fragment_shader *fs);
 
 
 
@@ -206,8 +210,5 @@ extern void i915_disassemble_program(const uint * program, uint sz);
 extern void
 i915_program_error(struct i915_fp_compile *p, const char *msg, ...);
 
-extern void
-i915_translate_fragment_program(struct i915_context *i915);
-
 
 #endif
diff --git a/src/gallium/drivers/i915simple/i915_fpc_emit.c b/src/gallium/drivers/i915simple/i915_fpc_emit.c
index 74924ff0a1d..4bdeefb449b 100644
--- a/src/gallium/drivers/i915simple/i915_fpc_emit.c
+++ b/src/gallium/drivers/i915simple/i915_fpc_emit.c
@@ -61,8 +61,6 @@
   			   (REG_NR_MASK << UREG_NR_SHIFT))
 
 
-#define I915_CONSTFLAG_PARAM 0x1f
-
 uint
 i915_get_temp(struct i915_fp_compile *p)
 {
@@ -73,10 +71,21 @@ i915_get_temp(struct i915_fp_compile *p)
    }
 
    p->temp_flag |= 1 << (bit - 1);
-   return UREG(REG_TYPE_R, (bit - 1));
+   return bit - 1;
+}
+
+
+static void
+i915_release_temp(struct i915_fp_compile *p, int reg)
+{
+   p->temp_flag &= ~(1 << reg);
 }
 
 
+/**
+ * Get unpreserved temporary, a temp whose value is not preserved between
+ * PS program phases.
+ */
 uint
 i915_get_utemp(struct i915_fp_compile * p)
 {
@@ -185,41 +194,62 @@ i915_emit_arith(struct i915_fp_compile * p,
    return dest;
 }
 
+
+/**
+ * Emit a texture load or texkill instruction.
+ * \param dest  the dest i915 register
+ * \param destmask  the dest register writemask
+ * \param sampler  the i915 sampler register
+ * \param coord  the i915 source texcoord operand
+ * \param opcode  the instruction opcode
+ */
 uint i915_emit_texld( struct i915_fp_compile *p,
 			uint dest,
 			uint destmask,
 			uint sampler,
 			uint coord,
-			uint op )
+			uint opcode )
 {
-   uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
+   const uint k = UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord));
+   int temp = -1;
+
    if (coord != k) {
-      /* No real way to work around this in the general case - need to
-       * allocate and declare a new temporary register (a utemp won't
-       * do).  Will fallback for now.
+      /* texcoord is swizzled or negated.  Need to allocate a new temporary
+       * register (a utemp / unpreserved temp) won't do.
        */
-      i915_program_error(p, "Can't (yet) swizzle TEX arguments");
-      assert(0);
-      return 0;
+      uint tempReg;
+
+      temp = i915_get_temp(p);           /* get temp reg index */
+      tempReg = UREG(REG_TYPE_R, temp);  /* make i915 register */
+
+      i915_emit_arith( p, A0_MOV,
+                       tempReg, A0_DEST_CHANNEL_ALL, /* dest reg, writemask */
+                       0,                            /* saturate */
+                       coord, 0, 0 );                /* src0, src1, src2 */
+
+      /* new src texcoord is tempReg */
+      coord = tempReg;
    }
 
    /* Don't worry about saturate as we only support  
     */
    if (destmask != A0_DEST_CHANNEL_ALL) {
+      /* if not writing to XYZW... */
       uint tmp = i915_get_utemp(p);
-      i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op );
+      i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, opcode );
       i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
-      return dest;
+      /* XXX release utemp here? */
    }
    else {
       assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
       assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
 
+      /* is the sampler coord a texcoord input reg? */
       if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
 	 p->nr_tex_indirect++;
       }
 
-      *(p->csr++) = (op | 
+      *(p->csr++) = (opcode | 
 		     T0_DEST( dest ) |
 		     T0_SAMPLER( sampler ));
 
@@ -227,14 +257,19 @@ uint i915_emit_texld( struct i915_fp_compile *p,
       *(p->csr++) = T2_MBZ;
 
       p->nr_tex_insn++;
-      return dest;
    }
+
+   if (temp >= 0)
+      i915_release_temp(p, temp);
+
+   return dest;
 }
 
 
 uint
 i915_emit_const1f(struct i915_fp_compile * p, float c0)
 {
+   struct i915_fragment_shader *ifs = p->shader;
    unsigned reg, idx;
 
    if (c0 == 0.0)
@@ -243,15 +278,15 @@ i915_emit_const1f(struct i915_fp_compile * p, float c0)
       return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
 
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
-      if (p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
+      if (ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
          continue;
       for (idx = 0; idx < 4; idx++) {
-         if (!(p->constant_flags[reg] & (1 << idx)) ||
-             p->constants[reg][idx] == c0) {
-            p->constants[reg][idx] = c0;
-            p->constant_flags[reg] |= 1 << idx;
-            if (reg + 1 > p->num_constants)
-               p->num_constants = reg + 1;
+         if (!(ifs->constant_flags[reg] & (1 << idx)) ||
+             ifs->constants[reg][idx] == c0) {
+            ifs->constants[reg][idx] = c0;
+            ifs->constant_flags[reg] |= 1 << idx;
+            if (reg + 1 > ifs->num_constants)
+               ifs->num_constants = reg + 1;
             return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
          }
       }
@@ -264,6 +299,7 @@ i915_emit_const1f(struct i915_fp_compile * p, float c0)
 uint
 i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
 {
+   struct i915_fragment_shader *ifs = p->shader;
    unsigned reg, idx;
 
    if (c0 == 0.0)
@@ -277,16 +313,16 @@ i915_emit_const2f(struct i915_fp_compile * p, float c0, float c1)
       return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
 
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
-      if (p->constant_flags[reg] == 0xf ||
-          p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
+      if (ifs->constant_flags[reg] == 0xf ||
+          ifs->constant_flags[reg] == I915_CONSTFLAG_USER)
          continue;
       for (idx = 0; idx < 3; idx++) {
-         if (!(p->constant_flags[reg] & (3 << idx))) {
-            p->constants[reg][idx + 0] = c0;
-            p->constants[reg][idx + 1] = c1;
-            p->constant_flags[reg] |= 3 << idx;
-            if (reg + 1 > p->num_constants)
-               p->num_constants = reg + 1;
+         if (!(ifs->constant_flags[reg] & (3 << idx))) {
+            ifs->constants[reg][idx + 0] = c0;
+            ifs->constants[reg][idx + 1] = c1;
+            ifs->constant_flags[reg] |= 3 << idx;
+            if (reg + 1 > ifs->num_constants)
+               ifs->num_constants = reg + 1;
             return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO, ONE);
          }
       }
@@ -302,25 +338,26 @@ uint
 i915_emit_const4f(struct i915_fp_compile * p,
                   float c0, float c1, float c2, float c3)
 {
+   struct i915_fragment_shader *ifs = p->shader;
    unsigned reg;
 
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
-      if (p->constant_flags[reg] == 0xf &&
-          p->constants[reg][0] == c0 &&
-          p->constants[reg][1] == c1 &&
-          p->constants[reg][2] == c2 &&
-          p->constants[reg][3] == c3) {
+      if (ifs->constant_flags[reg] == 0xf &&
+          ifs->constants[reg][0] == c0 &&
+          ifs->constants[reg][1] == c1 &&
+          ifs->constants[reg][2] == c2 &&
+          ifs->constants[reg][3] == c3) {
          return UREG(REG_TYPE_CONST, reg);
       }
-      else if (p->constant_flags[reg] == 0) {
-
-         p->constants[reg][0] = c0;
-         p->constants[reg][1] = c1;
-         p->constants[reg][2] = c2;
-         p->constants[reg][3] = c3;
-         p->constant_flags[reg] = 0xf;
-         if (reg + 1 > p->num_constants)
-            p->num_constants = reg + 1;
+      else if (ifs->constant_flags[reg] == 0) {
+
+         ifs->constants[reg][0] = c0;
+         ifs->constants[reg][1] = c1;
+         ifs->constants[reg][2] = c2;
+         ifs->constants[reg][3] = c3;
+         ifs->constant_flags[reg] = 0xf;
+         if (reg + 1 > ifs->num_constants)
+            ifs->num_constants = reg + 1;
          return UREG(REG_TYPE_CONST, reg);
       }
    }
@@ -335,41 +372,3 @@ i915_emit_const4fv(struct i915_fp_compile * p, const float * c)
 {
    return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
 }
-
-
-#if 00000/*UNUSED*/
-/* Reserve a slot in the constant file for a Mesa state parameter.
- * These will later need to be tracked on statechanges, but that is
- * done elsewhere.
- */
-uint
-i915_emit_param4fv(struct i915_fp_compile * p, const float * values)
-{
-   struct i915_fragment_program *fp = p->fp;
-   int i;
-
-   for (i = 0; i < fp->nr_params; i++) {
-      if (fp->param[i].values == values)
-         return UREG(REG_TYPE_CONST, fp->param[i].reg);
-   }
-
-   if (p->constants->nr_constants == I915_MAX_CONSTANT ||
-       fp->nr_params == I915_MAX_CONSTANT) {
-      i915_program_error(p, "i915_emit_param4fv: out of constants\n");
-      return 0;
-   }
-
-   {
-      int reg = p->constants->nr_constants++;
-      int i = fp->nr_params++;
-
-      assert (p->constant_flags[reg] == 0);
-      p->constant_flags[reg] = I915_CONSTFLAG_PARAM;
-
-      fp->param[i].values = values;
-      fp->param[i].reg = reg;
-
-      return UREG(REG_TYPE_CONST, reg);
-   }
-}
-#endif
diff --git a/src/gallium/drivers/i915simple/i915_fpc_translate.c b/src/gallium/drivers/i915simple/i915_fpc_translate.c
index 6c1524c768e..76a2184e9ab 100644
--- a/src/gallium/drivers/i915simple/i915_fpc_translate.c
+++ b/src/gallium/drivers/i915simple/i915_fpc_translate.c
@@ -34,6 +34,7 @@
 
 #include "pipe/p_shader_tokens.h"
 #include "tgsi/util/tgsi_parse.h"
+#include "tgsi/util/tgsi_dump.h"
 
 #include "draw/draw_vertex.h"
 
@@ -97,19 +98,19 @@ negate(int reg, int x, int y, int z, int w)
 }
 
 
+/**
+ * In the event of a translation failure, we'll generate a simple color
+ * pass-through program.
+ */
 static void
-i915_use_passthrough_shader(struct i915_context *i915)
+i915_use_passthrough_shader(struct i915_fragment_shader *fs)
 {
-   debug_printf("**** Using i915 pass-through fragment shader\n");
-
-   i915->current.program = (uint *) MALLOC(sizeof(passthrough));
-   if (i915->current.program) {
-      memcpy(i915->current.program, passthrough, sizeof(passthrough));
-      i915->current.program_len = Elements(passthrough);
+   fs->program = (uint *) MALLOC(sizeof(passthrough));
+   if (fs->program) {
+      memcpy(fs->program, passthrough, sizeof(passthrough));
+      fs->program_len = Elements(passthrough);
    }
-
-   i915->current.num_constants[PIPE_SHADER_FRAGMENT] = 0;
-   i915->current.num_user_constants[PIPE_SHADER_FRAGMENT] = 0;
+   fs->num_constants = 0;
 }
 
 
@@ -161,9 +162,6 @@ src_vector(struct i915_fp_compile *p,
        * We also use a texture coordinate to pass wpos when possible.
        */
 
-      /* use vertex format info to map a slot number to a VF attrib */
-      assert(index < p->vertex_info->num_attribs);
-
       sem_name = p->input_semantic_name[index];
       sem_ind = p->input_semantic_index[index];
 
@@ -201,7 +199,8 @@ src_vector(struct i915_fp_compile *p,
       break;
 
    case TGSI_FILE_IMMEDIATE:
-      /* XXX unfinished - need to append immediates onto const buffer */
+      assert(index < p->num_immediates);
+      index = p->immediates_map[index];
       /* fall-through */
    case TGSI_FILE_CONSTANT:
       src = UREG(REG_TYPE_CONST, index);
@@ -386,6 +385,26 @@ emit_simple_arith(struct i915_fp_compile *p,
                     arg3 );
 }
 
+
+/** As above, but swap the first two src regs */
+static void
+emit_simple_arith_swap2(struct i915_fp_compile *p,
+                        const struct tgsi_full_instruction *inst,
+                        uint opcode, uint numArgs)
+{
+   struct tgsi_full_instruction inst2;
+
+   assert(numArgs == 2);
+
+   /* transpose first two registers */
+   inst2 = *inst;
+   inst2.FullSrcRegisters[0] = inst->FullSrcRegisters[1];
+   inst2.FullSrcRegisters[1] = inst->FullSrcRegisters[0];
+
+   emit_simple_arith(p, &inst2, opcode, numArgs);
+}
+
+
 #ifndef M_PI
 #define M_PI 3.14159265358979323846
 #endif
@@ -556,8 +575,12 @@ i915_translate_instruction(struct i915_fp_compile *p,
       src0 = src_vector(p, &inst->FullSrcRegisters[0]);
       tmp = i915_get_utemp(p);
 
-      i915_emit_texld(p, tmp, A0_DEST_CHANNEL_ALL,   /* use a dummy dest reg */
-                      0, src0, T0_TEXKILL);
+      i915_emit_texld(p,
+                      tmp,                   /* dest reg: a dummy reg */
+                      A0_DEST_CHANNEL_ALL,   /* dest writemask */
+                      0,                     /* sampler */
+                      src0,                  /* coord*/
+                      T0_TEXKILL);           /* opcode */
       break;
 
    case TGSI_OPCODE_LG2:
@@ -773,6 +796,11 @@ i915_translate_instruction(struct i915_fp_compile *p,
       emit_simple_arith(p, inst, A0_SGE, 2);
       break;
 
+   case TGSI_OPCODE_SLE:
+      /* like SGE, but swap reg0, reg1 */
+      emit_simple_arith_swap2(p, inst, A0_SGE, 2);
+      break;
+
    case TGSI_OPCODE_SIN:
       src0 = src_vector(p, &inst->FullSrcRegisters[0]);
       tmp = i915_get_utemp(p);
@@ -827,6 +855,11 @@ i915_translate_instruction(struct i915_fp_compile *p,
       emit_simple_arith(p, inst, A0_SLT, 2);
       break;
 
+   case TGSI_OPCODE_SGT:
+      /* like SLT, but swap reg0, reg1 */
+      emit_simple_arith_swap2(p, inst, A0_SLT, 2);
+      break;
+
    case TGSI_OPCODE_SUB:
       src0 = src_vector(p, &inst->FullSrcRegisters[0]);
       src1 = src_vector(p, &inst->FullSrcRegisters[1]);
@@ -880,6 +913,7 @@ i915_translate_instruction(struct i915_fp_compile *p,
 
    default:
       i915_program_error(p, "bad opcode %d", inst->Instruction.Opcode);
+      p->error = 1;
       return;
    }
 
@@ -896,6 +930,7 @@ static void
 i915_translate_instructions(struct i915_fp_compile *p,
                             const struct tgsi_token *tokens)
 {
+   struct i915_fragment_shader *ifs = p->shader;
    struct tgsi_parse_context parse;
 
    tgsi_parse_init( &parse, tokens );
@@ -928,13 +963,64 @@ i915_translate_instructions(struct i915_fp_compile *p,
             p->output_semantic_name[ind] = sem;
             p->output_semantic_index[ind] = semi;
          }
+         else if (parse.FullToken.FullDeclaration.Declaration.File
+                  == TGSI_FILE_CONSTANT) {
+            uint i;
+            for (i = parse.FullToken.FullDeclaration.u.DeclarationRange.First;
+                 i <= parse.FullToken.FullDeclaration.u.DeclarationRange.Last;
+                 i++) {
+               assert(ifs->constant_flags[i] == 0x0);
+               ifs->constant_flags[i] = I915_CONSTFLAG_USER;
+               ifs->num_constants = MAX2(ifs->num_constants, i + 1);
+            }
+         }
+         else if (parse.FullToken.FullDeclaration.Declaration.File
+                  == TGSI_FILE_TEMPORARY) {
+            uint i;
+            for (i = parse.FullToken.FullDeclaration.u.DeclarationRange.First;
+                 i <= parse.FullToken.FullDeclaration.u.DeclarationRange.Last;
+                 i++) {
+               assert(i < I915_MAX_TEMPORARY);
+               p->temp_flag |= (1 << i); /* mark temp as used */
+            }
+         }
          break;
 
       case TGSI_TOKEN_TYPE_IMMEDIATE:
-         /* XXX append the immediate to the const buffer... */
+         {
+            const struct tgsi_full_immediate *imm
+               = &parse.FullToken.FullImmediate;
+            const uint pos = p->num_immediates++;
+            uint j;
+            for (j = 0; j < imm->Immediate.Size; j++) {
+               p->immediates[pos][j] = imm->u.ImmediateFloat32[j].Float;
+            }
+         }
          break;
 
       case TGSI_TOKEN_TYPE_INSTRUCTION:
+         if (p->first_instruction) {
+            /* resolve location of immediates */
+            uint i, j;
+            for (i = 0; i < p->num_immediates; i++) {
+               /* find constant slot for this immediate */
+               for (j = 0; j < I915_MAX_CONSTANT; j++) {
+                  if (ifs->constant_flags[j] == 0x0) {
+                     memcpy(ifs->constants[j],
+                            p->immediates[i],
+                            4 * sizeof(float));
+                     /*printf("immediate %d maps to const %d\n", i, j);*/
+                     ifs->constant_flags[j] = 0xf;  /* all four comps used */
+                     p->immediates_map[i] = j;
+                     ifs->num_constants = MAX2(ifs->num_constants, j + 1);
+                     break;
+                  }
+               }
+            }
+
+            p->first_instruction = FALSE;
+         }
+
          i915_translate_instruction(p, &parse.FullToken.FullInstruction);
          break;
 
@@ -950,32 +1036,33 @@ i915_translate_instructions(struct i915_fp_compile *p,
 
 static struct i915_fp_compile *
 i915_init_compile(struct i915_context *i915,
-                  const struct pipe_shader_state *fs)
+                  struct i915_fragment_shader *ifs)
 {
    struct i915_fp_compile *p = CALLOC_STRUCT(i915_fp_compile);
 
-   p->shader = i915->fs;
+   p->shader = ifs;
 
-   p->vertex_info = &i915->current.vertex_info;
-
-   /* new constants found during translation get appended after the
-    * user-provided constants.
+   /* Put new constants at end of const buffer, growing downward.
+    * The problem is we don't know how many user-defined constants might
+    * be specified with pipe->set_constant_buffer().
+    * Should pre-scan the user's program to determine the highest-numbered
+    * constant referenced.
     */
-   p->constants = i915->current.constants[PIPE_SHADER_FRAGMENT];
-   p->num_constants = i915->current.num_user_constants[PIPE_SHADER_FRAGMENT];
+   ifs->num_constants = 0;
+   memset(ifs->constant_flags, 0, sizeof(ifs->constant_flags));
+
+   p->first_instruction = TRUE;
 
    p->nr_tex_indirect = 1;      /* correct? */
    p->nr_tex_insn = 0;
    p->nr_alu_insn = 0;
    p->nr_decl_insn = 0;
 
-   memset(p->constant_flags, 0, sizeof(p->constant_flags));
-
    p->csr = p->program;
    p->decl = p->declarations;
    p->decl_s = 0;
    p->decl_t = 0;
-   p->temp_flag = 0xffff000;
+   p->temp_flag = ~0x0 << I915_MAX_TEMPORARY;
    p->utemp_flag = ~0x7;
 
    p->wpos_tex = -1;
@@ -993,6 +1080,7 @@ i915_init_compile(struct i915_context *i915,
 static void
 i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
 {
+   struct i915_fragment_shader *ifs = p->shader;
    unsigned long program_size = (unsigned long) (p->csr - p->program);
    unsigned long decl_size = (unsigned long) (p->decl - p->declarations);
 
@@ -1008,19 +1096,13 @@ i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
    if (p->nr_decl_insn > I915_MAX_DECL_INSN)
       i915_program_error(p, "Exceeded max DECL instructions");
 
-   /* free old program, if present */
-   if (i915->current.program) {
-      FREE(i915->current.program);
-      i915->current.program_len = 0;
-   }
-
    if (p->error) {
       p->NumNativeInstructions = 0;
       p->NumNativeAluInstructions = 0;
       p->NumNativeTexInstructions = 0;
       p->NumNativeTexIndirections = 0;
 
-      i915_use_passthrough_shader(i915);
+      i915_use_passthrough_shader(ifs);
    }
    else {
       p->NumNativeInstructions
@@ -1034,24 +1116,20 @@ i915_fini_compile(struct i915_context *i915, struct i915_fp_compile *p)
 
       /* Copy compilation results to fragment program struct: 
        */
-      i915->current.program
+      assert(!ifs->program);
+      ifs->program
          = (uint *) MALLOC((program_size + decl_size) * sizeof(uint));
-      if (i915->current.program) {
-         i915->current.program_len = program_size + decl_size;
+      if (ifs->program) {
+         ifs->program_len = program_size + decl_size;
 
-         memcpy(i915->current.program,
+         memcpy(ifs->program,
                 p->declarations, 
                 decl_size * sizeof(uint));
 
-         memcpy(i915->current.program + decl_size, 
+         memcpy(ifs->program + decl_size, 
                 p->program, 
                 program_size * sizeof(uint));
       }
-
-      /* update number of constants */
-      i915->current.num_constants[PIPE_SHADER_FRAGMENT] = p->num_constants;
-      assert(i915->current.num_constants[PIPE_SHADER_FRAGMENT]
-             >= i915->current.num_user_constants[PIPE_SHADER_FRAGMENT]);
    }
 
    /* Release the compilation struct: 
@@ -1085,7 +1163,7 @@ i915_find_wpos_space(struct i915_fp_compile *p)
       i915_program_error(p, "No free texcoord for wpos value");
    }
 #else
-   if (p->shader->input_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
+   if (p->shader->state.input_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
       /* frag shader using the fragment position input */
 #if 0
       assert(0);
@@ -1106,7 +1184,7 @@ static void
 i915_fixup_depth_write(struct i915_fp_compile *p)
 {
    /* XXX assuming pos/depth is always in output[0] */
-   if (p->shader->output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
+   if (p->shader->state.output_semantic_name[0] == TGSI_SEMANTIC_POSITION) {
       const uint depth = UREG(REG_TYPE_OD, 0);
 
       i915_emit_arith(p,
@@ -1121,13 +1199,18 @@ i915_fixup_depth_write(struct i915_fp_compile *p)
 
 
 void
-i915_translate_fragment_program( struct i915_context *i915 )
+i915_translate_fragment_program( struct i915_context *i915,
+                                 struct i915_fragment_shader *fs)
 {
-   struct i915_fp_compile *p = i915_init_compile(i915, i915->fs);
-   const struct tgsi_token *tokens = i915->fs->tokens;
+   struct i915_fp_compile *p = i915_init_compile(i915, fs);
+   const struct tgsi_token *tokens = fs->state.tokens;
 
    i915_find_wpos_space(p);
 
+#if 0
+   tgsi_dump(tokens, 0);
+#endif
+
    i915_translate_instructions(p, tokens);
    i915_fixup_depth_write(p);
 
diff --git a/src/gallium/drivers/i915simple/i915_prim_emit.c b/src/gallium/drivers/i915simple/i915_prim_emit.c
index 44c43259369..d8de5178f60 100644
--- a/src/gallium/drivers/i915simple/i915_prim_emit.c
+++ b/src/gallium/drivers/i915simple/i915_prim_emit.c
@@ -72,38 +72,42 @@ emit_hw_vertex( struct i915_context *i915,
    uint i;
    uint count = 0;  /* for debug/sanity */
 
+   assert(!i915->dirty);
+
    for (i = 0; i < vinfo->num_attribs; i++) {
+      const uint j = vinfo->src_index[i];
+      const float *attrib = vertex->data[j];
       switch (vinfo->emit[i]) {
       case EMIT_OMIT:
          /* no-op */
          break;
       case EMIT_1F:
-         OUT_BATCH( fui(vertex->data[i][0]) );
+         OUT_BATCH( fui(attrib[0]) );
          count++;
          break;
       case EMIT_2F:
-         OUT_BATCH( fui(vertex->data[i][0]) );
-         OUT_BATCH( fui(vertex->data[i][1]) );
+         OUT_BATCH( fui(attrib[0]) );
+         OUT_BATCH( fui(attrib[1]) );
          count += 2;
          break;
       case EMIT_3F:
-         OUT_BATCH( fui(vertex->data[i][0]) );
-         OUT_BATCH( fui(vertex->data[i][1]) );
-         OUT_BATCH( fui(vertex->data[i][2]) );
+         OUT_BATCH( fui(attrib[0]) );
+         OUT_BATCH( fui(attrib[1]) );
+         OUT_BATCH( fui(attrib[2]) );
          count += 3;
          break;
       case EMIT_4F:
-         OUT_BATCH( fui(vertex->data[i][0]) );
-         OUT_BATCH( fui(vertex->data[i][1]) );
-         OUT_BATCH( fui(vertex->data[i][2]) );
-         OUT_BATCH( fui(vertex->data[i][3]) );
+         OUT_BATCH( fui(attrib[0]) );
+         OUT_BATCH( fui(attrib[1]) );
+         OUT_BATCH( fui(attrib[2]) );
+         OUT_BATCH( fui(attrib[3]) );
          count += 4;
          break;
       case EMIT_4UB:
-         OUT_BATCH( pack_ub4(float_to_ubyte( vertex->data[i][2] ),
-                             float_to_ubyte( vertex->data[i][1] ),
-                             float_to_ubyte( vertex->data[i][0] ),
-                             float_to_ubyte( vertex->data[i][3] )) );
+         OUT_BATCH( pack_ub4(float_to_ubyte( attrib[2] ),
+                             float_to_ubyte( attrib[1] ),
+                             float_to_ubyte( attrib[0] ),
+                             float_to_ubyte( attrib[3] )) );
          count += 1;
          break;
       default:
@@ -122,17 +126,19 @@ emit_prim( struct draw_stage *stage,
 	   unsigned nr )
 {
    struct i915_context *i915 = setup_stage(stage)->i915;
-   unsigned vertex_size = i915->current.vertex_info.size * 4; /* in bytes */
+   unsigned vertex_size;
    unsigned i;
 
-   assert(vertex_size >= 12); /* never smaller than 12 bytes */
-
    if (i915->dirty)
       i915_update_derived( i915 );
 
    if (i915->hardware_dirty)
       i915_emit_hardware_state( i915 );
 
+   /* need to do this after validation! */
+   vertex_size = i915->current.vertex_info.size * 4; /* in bytes */
+   assert(vertex_size >= 12); /* never smaller than 12 bytes */
+
    if (!BEGIN_BATCH( 1 + nr * vertex_size / 4, 0 )) {
       FLUSH_BATCH();
 
diff --git a/src/gallium/drivers/i915simple/i915_prim_vbuf.c b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
index c5bf6174f68..9d5f609220a 100644
--- a/src/gallium/drivers/i915simple/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915simple/i915_prim_vbuf.c
@@ -83,6 +83,12 @@ i915_vbuf_render_get_vertex_info( struct vbuf_render *render )
 {
    struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
    struct i915_context *i915 = i915_render->i915;
+
+   if (i915->dirty) {
+      /* make sure we have up to date vertex layout */
+      i915_update_derived( i915 );
+   }
+
    return &i915->current.vertex_info;
 }
 
@@ -143,7 +149,8 @@ i915_vbuf_render_draw( struct vbuf_render *render,
 
    assert(nr_indices);
 
-   assert((i915->dirty & ~I915_NEW_VBO) == 0);
+   /* this seems to be bogus, since we validate state right after this */
+   /*assert((i915->dirty & ~I915_NEW_VBO) == 0);*/
    
    if (i915->dirty)
       i915_update_derived( i915 );
diff --git a/src/gallium/drivers/i915simple/i915_state.c b/src/gallium/drivers/i915simple/i915_state.c
index e055eed7e02..a35bdf941fc 100644
--- a/src/gallium/drivers/i915simple/i915_state.c
+++ b/src/gallium/drivers/i915simple/i915_state.c
@@ -38,6 +38,7 @@
 #include "i915_reg.h"
 #include "i915_state.h"
 #include "i915_state_inlines.h"
+#include "i915_fpc.h"
 
 
 /* The i915 (and related graphics cores) do not support GL_CLAMP.  The
@@ -416,26 +417,47 @@ static void i915_set_polygon_stipple( struct pipe_context *pipe,
 }
 
 
-static void * i915_create_fs_state(struct pipe_context *pipe,
-                                   const struct pipe_shader_state *templ)
+
+static void *
+i915_create_fs_state(struct pipe_context *pipe,
+                     const struct pipe_shader_state *templ)
 {
-   return 0;
+   struct i915_context *i915 = i915_context(pipe);
+   struct i915_fragment_shader *ifs = CALLOC_STRUCT(i915_fragment_shader);
+   if (!ifs)
+      return NULL;
+
+   ifs->state = *templ;
+
+   /* The shader's compiled to i915 instructions here */
+   i915_translate_fragment_program(i915, ifs);
+
+   return ifs;
 }
 
-static void i915_bind_fs_state(struct pipe_context *pipe, void *fs)
+static void
+i915_bind_fs_state(struct pipe_context *pipe, void *shader)
 {
    struct i915_context *i915 = i915_context(pipe);
 
-   i915->fs = (struct pipe_shader_state *)fs;
+   i915->fs = (struct i915_fragment_shader*) shader;
 
    i915->dirty |= I915_NEW_FS;
 }
 
-static void i915_delete_fs_state(struct pipe_context *pipe, void *shader)
+static
+void i915_delete_fs_state(struct pipe_context *pipe, void *shader)
 {
-   /*do nothing*/
+   struct i915_fragment_shader *ifs = (struct i915_fragment_shader *) shader;
+
+   if (ifs->program)
+      FREE(ifs->program);
+   ifs->program_len = 0;
+
+   FREE(ifs);
 }
 
+
 static void *
 i915_create_vs_state(struct pipe_context *pipe,
                      const struct pipe_shader_state *templ)
@@ -452,6 +474,8 @@ static void i915_bind_vs_state(struct pipe_context *pipe, void *shader)
 
    /* just pass-through to draw module */
    draw_bind_vertex_shader(i915->draw, (struct draw_vertex_shader *) shader);
+
+   i915->dirty |= I915_NEW_VS;
 }
 
 static void i915_delete_vs_state(struct pipe_context *pipe, void *shader)
diff --git a/src/gallium/drivers/i915simple/i915_state_derived.c b/src/gallium/drivers/i915simple/i915_state_derived.c
index 4767584fc60..5cf70acdf3b 100644
--- a/src/gallium/drivers/i915simple/i915_state_derived.c
+++ b/src/gallium/drivers/i915simple/i915_state_derived.c
@@ -27,104 +27,111 @@
 
 
 #include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
 #include "draw/draw_context.h"
 #include "draw/draw_vertex.h"
 #include "i915_context.h"
 #include "i915_state.h"
 #include "i915_reg.h"
 #include "i915_fpc.h"
-#include "pipe/p_shader_tokens.h"
+
 
 
 /**
- * Determine which post-transform / pre-rasterization vertex attributes
- * we need.
- * Derived from:  fs, setup states.
+ * Determine the hardware vertex layout.
+ * Depends on vertex/fragment shader state.
  */
 static void calculate_vertex_layout( struct i915_context *i915 )
 {
-   const struct pipe_shader_state *fs = i915->fs;
+   const struct pipe_shader_state *fs = &i915->fs->state;
    const enum interp_mode colorInterp = i915->rasterizer->color_interp;
    struct vertex_info vinfo;
-   uint front0 = 0, back0 = 0, front1 = 0, back1 = 0;
-   boolean needW = 0;
+   boolean texCoords[8], colors[2], fog, needW;
    uint i;
-   boolean texCoords[8];
-   uint src = 0;
+   int src;
 
    memset(texCoords, 0, sizeof(texCoords));
+   colors[0] = colors[1] = fog = needW = FALSE;
    memset(&vinfo, 0, sizeof(vinfo));
 
-   /* pos */
-   draw_emit_vertex_attr(&vinfo, EMIT_3F, INTERP_LINEAR, src++);
-   /* Note: we'll set the S4_VFMT_XYZ[W] bits below */
-
+   /* Determine which fragment program inputs are needed.  Setup HW vertex
+    * layout below, in the HW-specific attribute order.
+    */
    for (i = 0; i < fs->num_inputs; i++) {
       switch (fs->input_semantic_name[i]) {
       case TGSI_SEMANTIC_POSITION:
          break;
       case TGSI_SEMANTIC_COLOR:
-         if (fs->input_semantic_index[i] == 0) {
-            front0 = draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src++);
-            vinfo.hwfmt[0] |= S4_VFMT_COLOR;
-         }
-         else {
-            assert(fs->input_semantic_index[i] == 1);
-            front1 = draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src++);
-            vinfo.hwfmt[0] |= S4_VFMT_SPEC_FOG;
-         }
+         assert(fs->input_semantic_index[i] < 2);
+         colors[fs->input_semantic_index[i]] = TRUE;
          break;
       case TGSI_SEMANTIC_GENERIC:
          /* usually a texcoord */
          {
             const uint unit = fs->input_semantic_index[i];
-            uint hwtc;
+            assert(unit < 8);
             texCoords[unit] = TRUE;
-            draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE, src++);
-            hwtc = TEXCOORDFMT_4D;
             needW = TRUE;
-            vinfo.hwfmt[1] |= hwtc << (unit * 4);
          }
          break;
       case TGSI_SEMANTIC_FOG:
-         debug_printf("i915 fogcoord not implemented yet\n");
-         draw_emit_vertex_attr(&vinfo, EMIT_1F, INTERP_PERSPECTIVE, src++);
+         fog = TRUE;
          break;
       default:
          assert(0);
       }
-
    }
 
-   /* finish up texcoord fields */
-   for (i = 0; i < 8; i++) {
-      if (!texCoords[i]) {
-         const uint hwtc = TEXCOORDFMT_NOT_PRESENT;
-         vinfo.hwfmt[1] |= hwtc << (i* 4);
-      }
-   }
-
-   /* go back and fill in the vertex position info now that we have needW */
+   
+   /* pos */
+   src = draw_find_vs_output(i915->draw, TGSI_SEMANTIC_POSITION, 0);
    if (needW) {
+      draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_LINEAR, src);
       vinfo.hwfmt[0] |= S4_VFMT_XYZW;
       vinfo.emit[0] = EMIT_4F;
    }
    else {
+      draw_emit_vertex_attr(&vinfo, EMIT_3F, INTERP_LINEAR, src);
       vinfo.hwfmt[0] |= S4_VFMT_XYZ;
       vinfo.emit[0] = EMIT_3F;
    }
 
-   /* Additional attributes required for setup: Just twosided
-    * lighting.  Edgeflag is dealt with specially by setting bits in
-    * the vertex header.
-    */
-   if (i915->rasterizer->light_twoside) {
-      if (front0) {
-         back0 = draw_emit_vertex_attr(&vinfo, EMIT_OMIT, colorInterp, src++);
+   /* hardware point size */
+   /* XXX todo */
+
+   /* primary color */
+   if (colors[0]) {
+      src = draw_find_vs_output(i915->draw, TGSI_SEMANTIC_COLOR, 0);
+      draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src);
+      vinfo.hwfmt[0] |= S4_VFMT_COLOR;
+   }
+
+   /* secondary color */
+   if (colors[1]) {
+      src = draw_find_vs_output(i915->draw, TGSI_SEMANTIC_COLOR, 1);
+      draw_emit_vertex_attr(&vinfo, EMIT_4UB, colorInterp, src);
+      vinfo.hwfmt[0] |= S4_VFMT_SPEC_FOG;
+   }
+
+   /* fog coord, not fog blend factor */
+   if (fog) {
+      src = draw_find_vs_output(i915->draw, TGSI_SEMANTIC_FOG, 0);
+      draw_emit_vertex_attr(&vinfo, EMIT_1F, INTERP_PERSPECTIVE, src);
+      vinfo.hwfmt[0] |= S4_VFMT_FOG_PARAM;
+   }
+
+   /* texcoords */
+   for (i = 0; i < 8; i++) {
+      uint hwtc;
+      if (texCoords[i]) {
+         hwtc = TEXCOORDFMT_4D;
+         src = draw_find_vs_output(i915->draw, TGSI_SEMANTIC_GENERIC, i);
+         draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
       }
-      if (back0) {
-         back1 = draw_emit_vertex_attr(&vinfo, EMIT_OMIT, colorInterp, src++);
+      else {
+         hwtc = TEXCOORDFMT_NOT_PRESENT;
       }
+      vinfo.hwfmt[1] |= hwtc << (i * 4);
    }
 
    draw_compute_vertex_size(&vinfo);
@@ -148,7 +155,7 @@ static void calculate_vertex_layout( struct i915_context *i915 )
  */
 void i915_update_derived( struct i915_context *i915 )
 {
-   if (i915->dirty & (I915_NEW_RASTERIZER | I915_NEW_FS))
+   if (i915->dirty & (I915_NEW_RASTERIZER | I915_NEW_FS | I915_NEW_VS))
       calculate_vertex_layout( i915 );
 
    if (i915->dirty & (I915_NEW_SAMPLER | I915_NEW_TEXTURE))
@@ -164,7 +171,6 @@ void i915_update_derived( struct i915_context *i915 )
       i915_update_dynamic( i915 );
 
    if (i915->dirty & I915_NEW_FS) {
-      i915_translate_fragment_program(i915);
       i915->hardware_dirty |= I915_HW_PROGRAM; /* XXX right? */
    }
 
diff --git a/src/gallium/drivers/i915simple/i915_state_emit.c b/src/gallium/drivers/i915simple/i915_state_emit.c
index 3339287f498..6bbaac4e34c 100644
--- a/src/gallium/drivers/i915simple/i915_state_emit.c
+++ b/src/gallium/drivers/i915simple/i915_state_emit.c
@@ -99,7 +99,11 @@ i915_emit_hardware_state(struct i915_context *i915 )
                              2 + I915_TEX_UNITS*3 + 
                              2 + I915_TEX_UNITS*3 +
                              2 + I915_MAX_CONSTANT*4 + 
+#if 0
                              i915->current.program_len + 
+#else
+                             i915->fs->program_len + 
+#endif
                              6 
                            ) * 3/2; /* plus 50% margin */
    const unsigned relocs = ( I915_TEX_UNITS +
@@ -325,15 +329,34 @@ i915_emit_hardware_state(struct i915_context *i915 )
    /* 2 + I915_MAX_CONSTANT*4 dwords, 0 relocs */
    if (i915->hardware_dirty & I915_HW_PROGRAM)
    {
-      const uint nr = i915->current.num_constants[PIPE_SHADER_FRAGMENT];
-      assert(nr <= I915_MAX_CONSTANT);
-      if (nr > 0) {
-         const uint *c
-            = (const uint *) i915->current.constants[PIPE_SHADER_FRAGMENT];
+      /* Collate the user-defined constants with the fragment shader's
+       * immediates according to the constant_flags[] array.
+       */
+      const uint nr = i915->fs->num_constants;
+      if (nr) {
          uint i;
+
          OUT_BATCH( _3DSTATE_PIXEL_SHADER_CONSTANTS | (nr * 4) );
          OUT_BATCH( (1 << (nr - 1)) | ((1 << (nr - 1)) - 1) );
+
          for (i = 0; i < nr; i++) {
+            const uint *c;
+            if (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER) {
+               /* grab user-defined constant */
+               c = (uint *) i915->current.constants[PIPE_SHADER_FRAGMENT][i];
+            }
+            else {
+               /* emit program constant */
+               c = (uint *) i915->fs->constants[i];
+            }
+#if 0 /* debug */
+            {
+               float *f = (float *) c;
+               printf("Const %2d: %f %f %f %f %s\n", i, f[0], f[1], f[2], f[3],
+                      (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER
+                       ? "user" : "immediate"));
+            }
+#endif
             OUT_BATCH(*c++);
             OUT_BATCH(*c++);
             OUT_BATCH(*c++);
@@ -348,9 +371,9 @@ i915_emit_hardware_state(struct i915_context *i915 )
    {
       uint i;
       /* we should always have, at least, a pass-through program */
-      assert(i915->current.program_len > 0);
-      for (i = 0; i < i915->current.program_len; i++) {
-         OUT_BATCH(i915->current.program[i]);
+      assert(i915->fs->program_len > 0);
+      for (i = 0; i < i915->fs->program_len; i++) {
+         OUT_BATCH(i915->fs->program[i]);
       }
    }
 
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
index d581ee8d3ca..4c1a6d5df0b 100644
--- a/src/gallium/drivers/softpipe/SConscript
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -5,6 +5,9 @@ env = env.Clone()
 softpipe = env.ConvenienceLibrary(
 	target = 'softpipe',
 	source = [
+		'sp_fs_exec.c',
+		'sp_fs_sse.c',
+		'sp_fs_llvm.c',
 		'sp_clear.c',
 		'sp_context.c',
 		'sp_draw_arrays.c',
diff --git a/src/gallium/drivers/softpipe/sp_fs_exec.c b/src/gallium/drivers/softpipe/sp_fs_exec.c
index 8cb0534342d..d5bd7a702f1 100644
--- a/src/gallium/drivers/softpipe/sp_fs_exec.c
+++ b/src/gallium/drivers/softpipe/sp_fs_exec.c
@@ -81,7 +81,7 @@ sp_setup_pos_vector(const struct tgsi_interp_coef *coef,
 
 
 static void
-exec_prepare( struct sp_fragment_shader *base,
+exec_prepare( const struct sp_fragment_shader *base,
 	      struct tgsi_exec_machine *machine,
 	      struct tgsi_sampler *samplers )
 {
@@ -98,7 +98,7 @@ exec_prepare( struct sp_fragment_shader *base,
  * interface:
  */
 static unsigned 
-exec_run( struct sp_fragment_shader *base,
+exec_run( const struct sp_fragment_shader *base,
 	  struct tgsi_exec_machine *machine,
 	  struct quad_header *quad )
 {
diff --git a/src/gallium/drivers/softpipe/sp_fs_llvm.c b/src/gallium/drivers/softpipe/sp_fs_llvm.c
index 22da4714533..34b2b7d4e24 100644
--- a/src/gallium/drivers/softpipe/sp_fs_llvm.c
+++ b/src/gallium/drivers/softpipe/sp_fs_llvm.c
@@ -146,7 +146,7 @@ shade_quad_llvm(struct quad_stage *qs,
 
 
 unsigned 
-run_llvm_fs( struct sp_fragment_shader *base,
+run_llvm_fs( const struct sp_fragment_shader *base,
 	     struct foo *machine )
 {
 }
diff --git a/src/gallium/drivers/softpipe/sp_prim_setup.c b/src/gallium/drivers/softpipe/sp_prim_setup.c
index 7b1e131ee14..b6a3fddb29c 100644
--- a/src/gallium/drivers/softpipe/sp_prim_setup.c
+++ b/src/gallium/drivers/softpipe/sp_prim_setup.c
@@ -1165,6 +1165,10 @@ static void setup_begin( struct draw_stage *stage )
    struct softpipe_context *sp = setup->softpipe;
    const struct pipe_shader_state *fs = &setup->softpipe->fs->shader;
 
+   if (sp->dirty) {
+      softpipe_update_derived(sp);
+   }
+
    setup->quad.nr_attrs = fs->num_inputs;
 
    sp->quad.first->begin(sp->quad.first);
diff --git a/src/gallium/drivers/softpipe/sp_quad.c b/src/gallium/drivers/softpipe/sp_quad.c
index 15b5594547d..142dbcc7710 100644
--- a/src/gallium/drivers/softpipe/sp_quad.c
+++ b/src/gallium/drivers/softpipe/sp_quad.c
@@ -56,11 +56,12 @@ sp_build_depth_stencil(
 void
 sp_build_quad_pipeline(struct softpipe_context *sp)
 {
-   boolean  early_depth_test =
+   boolean early_depth_test =
                sp->depth_stencil->depth.enabled &&
                sp->framebuffer.zsbuf &&
                !sp->depth_stencil->alpha.enabled &&
-               sp->fs->shader.output_semantic_name[0] != TGSI_SEMANTIC_POSITION;
+               !sp->fs->uses_kill &&
+               !sp->fs->writes_z;
 
    /* build up the pipeline in reverse order... */
 
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index ef8cf67d4c3..5aaa9e346bc 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -63,14 +63,17 @@ struct tgsi_exec_machine;
 struct sp_fragment_shader {
    struct pipe_shader_state   shader;
 
-   void (*prepare)( struct sp_fragment_shader *shader,
+   boolean uses_kill;
+   boolean writes_z;
+
+   void (*prepare)( const struct sp_fragment_shader *shader,
 		    struct tgsi_exec_machine *machine,
 		    struct tgsi_sampler *samplers);
 
    /* Run the shader - this interface will get cleaned up in the
     * future:
     */
-   unsigned (*run)( struct sp_fragment_shader *shader,
+   unsigned (*run)( const struct sp_fragment_shader *shader,
 		    struct tgsi_exec_machine *machine,
 		    struct quad_header *quad );
 
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
index f9f2c5eaa8f..4c6313001f4 100644
--- a/src/gallium/drivers/softpipe/sp_state_derived.c
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -35,33 +35,6 @@
 
 
 /**
- * Search vertex program's outputs to find a match for the given
- * semantic name/index.  Return the index of the output slot.
- *
- * Return 0 if not found.  This will cause the fragment program to use
- * vertex attrib 0 (position) in the cases where the fragment program
- * attempts to use a missing vertex program output.  This is an undefined
- * condition that users shouldn't hit anyway.
- */
-static int
-find_vs_output(struct softpipe_context *sp,
-               const struct pipe_shader_state *vs,
-               uint semantic_name,
-               uint semantic_index)
-{
-   uint i;
-   for (i = 0; i < vs->num_outputs; i++) {
-      if (vs->output_semantic_name[i] == semantic_name &&
-          vs->output_semantic_index[i] == semantic_index)
-         return i;
-   }
-
-   /* See if the draw module is introducing a new attribute... */
-   return draw_find_vs_output(sp->draw, semantic_name, semantic_index);
-}
-
-
-/**
  * Mark the current vertex layout as "invalid".
  * We'll validate the vertex layout later, when we start to actually
  * render a point or line or tri.
@@ -114,24 +87,25 @@ softpipe_get_vertex_info(struct softpipe_context *softpipe)
          int src;
          switch (fs->input_semantic_name[i]) {
          case TGSI_SEMANTIC_POSITION:
-            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_POSITION, 0);
+            src = draw_find_vs_output(softpipe->draw,
+                                      TGSI_SEMANTIC_POSITION, 0);
             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_POS, src);
             break;
 
          case TGSI_SEMANTIC_COLOR:
-            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_COLOR, 
+            src = draw_find_vs_output(softpipe->draw, TGSI_SEMANTIC_COLOR, 
                                  fs->input_semantic_index[i]);
             draw_emit_vertex_attr(vinfo, EMIT_4F, colorInterp, src);
             break;
 
          case TGSI_SEMANTIC_FOG:
-            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_FOG, 0);
+            src = draw_find_vs_output(softpipe->draw, TGSI_SEMANTIC_FOG, 0);
             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
             break;
 
          case TGSI_SEMANTIC_GENERIC:
             /* this includes texcoords and varying vars */
-            src = find_vs_output(softpipe, vs, TGSI_SEMANTIC_GENERIC,
+            src = draw_find_vs_output(softpipe->draw, TGSI_SEMANTIC_GENERIC,
                                  fs->input_semantic_index[i]);
             draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, src);
             break;
@@ -141,7 +115,8 @@ softpipe_get_vertex_info(struct softpipe_context *softpipe)
          }
       }
 
-      softpipe->psize_slot = find_vs_output(softpipe, vs, TGSI_SEMANTIC_PSIZE, 0);
+      softpipe->psize_slot = draw_find_vs_output(softpipe->draw,
+                                                 TGSI_SEMANTIC_PSIZE, 0);
       if (softpipe->psize_slot > 0) {
          draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_CONSTANT,
                                softpipe->psize_slot);
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index b0238f81737..b184ac61bb9 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -36,6 +36,7 @@
 #include "pipe/p_shader_tokens.h"
 #include "draw/draw_context.h"
 #include "tgsi/util/tgsi_dump.h"
+#include "tgsi/util/tgsi_scan.h"
 
 
 void *
@@ -44,21 +45,24 @@ softpipe_create_fs_state(struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
    struct sp_fragment_shader *state;
+   struct tgsi_shader_info info;
+
+   tgsi_scan_shader(templ->tokens, &info);
 
    if (softpipe->dump_fs) 
       tgsi_dump(templ->tokens, 0);
 
    state = softpipe_create_fs_llvm( softpipe, templ );
-   if (state)
-      return state;
-   
-   state = softpipe_create_fs_sse( softpipe, templ );
-   if (state)
-      return state;
-
-   state = softpipe_create_fs_exec( softpipe, templ );
-
+   if (!state) {
+      state = softpipe_create_fs_sse( softpipe, templ );
+      if (!state) {
+         state = softpipe_create_fs_exec( softpipe, templ );
+      }
+   }
    assert(state);
+   state->uses_kill = (info.opcode_count[TGSI_OPCODE_KIL] ||
+                       info.opcode_count[TGSI_OPCODE_KILP]);
+   state->writes_z = info.writes_z;
    return state;
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index 43d5085895f..0ced585c7f3 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -343,7 +343,7 @@ nearest_texcoord_unnorm(unsigned wrapMode, float s, unsigned size)
    switch (wrapMode) {
    case PIPE_TEX_WRAP_CLAMP:
       i = ifloor(s);
-      return CLAMP(i, 0, size-1);
+      return CLAMP(i, 0, (int) size-1);
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
       /* fall-through */
    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
@@ -366,7 +366,7 @@ linear_texcoord_unnorm(unsigned wrapMode, float s, unsigned size,
    switch (wrapMode) {
    case PIPE_TEX_WRAP_CLAMP:
       /* Not exactly what the spec says, but it matches NVIDIA output */
-      s = CLAMP(s - 0.5F, 0.0, (float) size - 1.0);
+      s = CLAMP(s - 0.5F, 0.0f, (float) size - 1.0f);
       *i0 = ifloor(s);
       *i1 = *i0 + 1;
       break;
@@ -377,7 +377,7 @@ linear_texcoord_unnorm(unsigned wrapMode, float s, unsigned size,
       s -= 0.5F;
       *i0 = ifloor(s);
       *i1 = *i0 + 1;
-      if (*i1 > size - 1)
+      if (*i1 > (int) size - 1)
          *i1 = size - 1;
       break;
    default: