Merge remote branch 'origin/master' into pipe-video

Conflicts:
	src/gallium/drivers/r600/r600_texture.c

11 files changed, 126 insertions, 56 deletions

diff --git a/src/gallium/drivers/i915/i915_context.c b/src/gallium/drivers/i915/i915_context.c
index 9be31619254..ea3c10b5e78 100644
--- a/src/gallium/drivers/i915/i915_context.c
+++ b/src/gallium/drivers/i915/i915_context.c
@@ -51,6 +51,14 @@ i915_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
    struct draw_context *draw = i915->draw;
    void *mapped_indices = NULL;
    unsigned i;
+   unsigned cbuf_dirty;
+
+
+   /*
+    * Ack vs contants here, helps ipers a lot.
+    */
+   cbuf_dirty = i915->dirty & I915_NEW_VS_CONSTANTS;
+   i915->dirty &= ~I915_NEW_VS_CONSTANTS;
 
    if (i915->dirty)
       i915_update_derived(i915);
@@ -70,10 +78,12 @@ i915_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
       mapped_indices = i915_buffer(i915->index_buffer.buffer)->data;
    draw_set_mapped_index_buffer(draw, mapped_indices);
 
-   draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX, 0,
-                                   i915->current.constants[PIPE_SHADER_VERTEX],
-                                   (i915->current.num_user_constants[PIPE_SHADER_VERTEX] * 
-                                      4 * sizeof(float)));
+   if (cbuf_dirty) {
+      draw_set_mapped_constant_buffer(draw, PIPE_SHADER_VERTEX, 0,
+                                      i915_buffer(i915->constants[PIPE_SHADER_VERTEX])->data,
+                                      (i915->current.num_user_constants[PIPE_SHADER_VERTEX] * 
+                                         4 * sizeof(float)));
+   }
 
    /*
     * Do the drawing
@@ -117,6 +127,11 @@ static void i915_destroy(struct pipe_context *pipe)
    }
    pipe_surface_reference(&i915->framebuffer.zsbuf, NULL);
 
+   /* unbind constant buffers */
+   for (i = 0; i < PIPE_SHADER_TYPES; i++) {
+      pipe_resource_reference(&i915->constants[i], NULL);
+   }
+
    FREE(i915);
 }
 
@@ -140,6 +155,10 @@ i915_create_context(struct pipe_screen *screen, void *priv)
 
    i915->base.draw_vbo = i915_draw_vbo;
 
+   /* init this before draw */
+   util_slab_create(&i915->transfer_pool, sizeof(struct pipe_transfer),
+                    16, UTIL_SLAB_SINGLETHREADED);
+
    /*
     * Create drawing context and plug our rendering stage into it.
     */
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index d15e1723d83..b7f1fb22221 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -37,6 +37,8 @@
 
 #include "tgsi/tgsi_scan.h"
 
+#include "util/u_slab.h"
+
 
 struct i915_winsys;
 struct i915_winsys_buffer;
@@ -134,7 +136,6 @@ struct i915_state
    unsigned immediate[I915_MAX_IMMEDIATE];
    unsigned dynamic[I915_MAX_DYNAMIC];
 
-   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];
 
@@ -212,7 +213,6 @@ struct i915_context {
    struct pipe_blend_color blend_color;
    struct pipe_stencil_ref stencil_ref;
    struct pipe_clip_state clip;
-   /* XXX unneded */
    struct pipe_resource *constants[PIPE_SHADER_TYPES];
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
@@ -237,6 +237,8 @@ struct i915_context {
 
    struct i915_state current;
    unsigned hardware_dirty;
+
+   struct util_slab_mempool transfer_pool;
 };
 
 /* A flag for each state_tracker state object:
@@ -253,9 +255,11 @@ struct i915_context {
 #define I915_NEW_DEPTH_STENCIL 0x200
 #define I915_NEW_SAMPLER       0x400
 #define I915_NEW_SAMPLER_VIEW  0x800
-#define I915_NEW_CONSTANTS     0x1000
-#define I915_NEW_VBO           0x2000
-#define I915_NEW_VS            0x4000
+#define I915_NEW_VS_CONSTANTS  0x1000
+#define I915_NEW_FS_CONSTANTS  0x2000
+#define I915_NEW_GS_CONSTANTS  0x4000
+#define I915_NEW_VBO           0x8000
+#define I915_NEW_VS            0x10000
 
 
 /* Driver's internally generated state flags:
diff --git a/src/gallium/drivers/i915/i915_debug.c b/src/gallium/drivers/i915/i915_debug.c
index 87c435a2f36..845e92cf5c6 100644
--- a/src/gallium/drivers/i915/i915_debug.c
+++ b/src/gallium/drivers/i915/i915_debug.c
@@ -948,7 +948,8 @@ i915_dump_dirty(struct i915_context *i915, const char *func)
       {I915_NEW_DEPTH_STENCIL, "depth_stencil"},
       {I915_NEW_SAMPLER,       "sampler"},
       {I915_NEW_SAMPLER_VIEW,  "sampler_view"},
-      {I915_NEW_CONSTANTS,     "constants"},
+      {I915_NEW_VS_CONSTANTS,  "vs_const"},
+      {I915_NEW_FS_CONSTANTS,  "fs_const"},
       {I915_NEW_VBO,           "vbo"},
       {I915_NEW_VS,            "vs"},
       {0, NULL},
diff --git a/src/gallium/drivers/i915/i915_prim_vbuf.c b/src/gallium/drivers/i915/i915_prim_vbuf.c
index baebbc7bae3..3473c863970 100644
--- a/src/gallium/drivers/i915/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915/i915_prim_vbuf.c
@@ -181,6 +181,7 @@ i915_vbuf_render_new_buf(struct i915_vbuf_render *i915_render, size_t size)
    struct i915_winsys *iws = i915->iws;
 
    if (i915_render->vbo) {
+      iws->buffer_unmap(iws, i915_render->vbo);
       iws->buffer_destroy(iws, i915_render->vbo);
       /*
        * XXX If buffers where referenced then this should be done in
@@ -208,6 +209,7 @@ i915_vbuf_render_new_buf(struct i915_vbuf_render *i915_render, size_t size)
 
    i915_render->vbo = iws->buffer_create(iws, i915_render->vbo_size,
                                          I915_NEW_VERTEX);
+   i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE);
 }
 
 /**
@@ -262,16 +264,13 @@ i915_vbuf_render_map_vertices(struct vbuf_render *render)
 {
    struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
    struct i915_context *i915 = i915_render->i915;
-   struct i915_winsys *iws = i915->iws;
 
    if (i915->vbo_flushed)
       debug_printf("%s bad vbo flush occured stalling on hw\n", __FUNCTION__);
 
 #ifdef VBUF_MAP_BUFFER
-   i915_render->vbo_ptr = iws->buffer_map(iws, i915_render->vbo, TRUE);
    return (unsigned char *)i915_render->vbo_ptr + i915_render->vbo_sw_offset;
 #else
-   (void)iws;
    return (unsigned char *)i915_render->vbo_ptr;
 #endif
 }
@@ -288,7 +287,7 @@ i915_vbuf_render_unmap_vertices(struct vbuf_render *render,
    i915_render->vbo_max_index = max_index;
    i915_render->vbo_max_used = MAX2(i915_render->vbo_max_used, i915_render->vertex_size * (max_index + 1));
 #ifdef VBUF_MAP_BUFFER
-   iws->buffer_unmap(iws, i915_render->vbo);
+   (void)iws;
 #else
    i915_render->map_used_start = i915_render->vertex_size * min_index;
    i915_render->map_used_end = i915_render->vertex_size * (max_index + 1);
@@ -684,6 +683,15 @@ static void
 i915_vbuf_render_destroy(struct vbuf_render *render)
 {
    struct i915_vbuf_render *i915_render = i915_vbuf_render(render);
+   struct i915_context *i915 = i915_render->i915;
+   struct i915_winsys *iws = i915->iws;
+
+   if (i915_render->vbo) {
+      i915->vbo = NULL;
+      iws->buffer_unmap(iws, i915_render->vbo);
+      iws->buffer_destroy(iws, i915_render->vbo);
+   }
+
    FREE(i915_render);
 }
 
diff --git a/src/gallium/drivers/i915/i915_resource_buffer.c b/src/gallium/drivers/i915/i915_resource_buffer.c
index 450203d60a9..51482f54fc4 100644
--- a/src/gallium/drivers/i915/i915_resource_buffer.c
+++ b/src/gallium/drivers/i915/i915_resource_buffer.c
@@ -60,6 +60,38 @@ i915_buffer_destroy(struct pipe_screen *screen,
 }
 
 
+static struct pipe_transfer *
+i915_get_transfer(struct pipe_context *pipe,
+                  struct pipe_resource *resource,
+                  unsigned level,
+                  unsigned usage,
+                  const struct pipe_box *box)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   struct pipe_transfer *transfer = util_slab_alloc(&i915->transfer_pool);
+
+   if (transfer == NULL)
+      return NULL;
+
+   transfer->resource = resource;
+   transfer->level = level;
+   transfer->usage = usage;
+   transfer->box = *box;
+
+   /* Note strides are zero, this is ok for buffers, but not for
+    * textures 2d & higher at least. 
+    */
+   return transfer;
+}
+
+static void
+i915_transfer_destroy(struct pipe_context *pipe,
+                      struct pipe_transfer *transfer)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   util_slab_free(&i915->transfer_pool, transfer);
+}
+
 static void *
 i915_buffer_transfer_map( struct pipe_context *pipe,
                           struct pipe_transfer *transfer )
@@ -92,8 +124,8 @@ struct u_resource_vtbl i915_buffer_vtbl =
    i915_buffer_get_handle,	     /* get_handle */
    i915_buffer_destroy,		     /* resource_destroy */
    NULL,			     /* is_resource_referenced */
-   u_default_get_transfer,	     /* get_transfer */
-   u_default_transfer_destroy,	     /* transfer_destroy */
+   i915_get_transfer,		     /* get_transfer */
+   i915_transfer_destroy,	     /* transfer_destroy */
    i915_buffer_transfer_map,	     /* transfer_map */
    u_default_transfer_flush_region,  /* transfer_flush_region */
    u_default_transfer_unmap,	     /* transfer_unmap */
@@ -115,8 +147,7 @@ i915_buffer_create(struct pipe_screen *screen,
    buf->b.vtbl = &i915_buffer_vtbl;
    pipe_reference_init(&buf->b.b.reference, 1);
    buf->b.b.screen = screen;
-
-   buf->data = MALLOC(template->width0);
+   buf->data = align_malloc(template->width0, 16);
    buf->free_on_destroy = TRUE;
 
    if (!buf->data)
diff --git a/src/gallium/drivers/i915/i915_resource_texture.c b/src/gallium/drivers/i915/i915_resource_texture.c
index f19106f3414..e793d126ade 100644
--- a/src/gallium/drivers/i915/i915_resource_texture.c
+++ b/src/gallium/drivers/i915/i915_resource_texture.c
@@ -716,14 +716,16 @@ i915_texture_destroy(struct pipe_screen *screen,
 }
 
 static struct pipe_transfer * 
-i915_texture_get_transfer(struct pipe_context *context,
+i915_texture_get_transfer(struct pipe_context *pipe,
                           struct pipe_resource *resource,
                           unsigned level,
                           unsigned usage,
                           const struct pipe_box *box)
 {
+   struct i915_context *i915 = i915_context(pipe);
    struct i915_texture *tex = i915_texture(resource);
-   struct pipe_transfer *transfer = CALLOC_STRUCT(pipe_transfer);
+   struct pipe_transfer *transfer = util_slab_alloc(&i915->transfer_pool);
+
    if (transfer == NULL)
       return NULL;
 
@@ -737,6 +739,14 @@ i915_texture_get_transfer(struct pipe_context *context,
    return transfer;
 }
 
+static void
+i915_transfer_destroy(struct pipe_context *pipe,
+                      struct pipe_transfer *transfer)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   util_slab_free(&i915->transfer_pool, transfer);
+}
+
 static void *
 i915_texture_transfer_map(struct pipe_context *pipe,
                           struct pipe_transfer *transfer)
@@ -781,7 +791,7 @@ struct u_resource_vtbl i915_texture_vtbl =
    i915_texture_destroy,	      /* resource_destroy */
    NULL,			      /* is_resource_referenced */
    i915_texture_get_transfer,	      /* get_transfer */
-   u_default_transfer_destroy,	      /* transfer_destroy */
+   i915_transfer_destroy,	      /* transfer_destroy */
    i915_texture_transfer_map,	      /* transfer_map */
    u_default_transfer_flush_region,   /* transfer_flush_region */
    i915_texture_transfer_unmap,	      /* transfer_unmap */
diff --git a/src/gallium/drivers/i915/i915_screen.h b/src/gallium/drivers/i915/i915_screen.h
index 0c4186c68ee..bb4d255a3b3 100644
--- a/src/gallium/drivers/i915/i915_screen.h
+++ b/src/gallium/drivers/i915/i915_screen.h
@@ -47,16 +47,6 @@ struct i915_screen
    boolean is_i945;
 };
 
-/**
- * Subclass of pipe_transfer
- */
-struct i915_transfer
-{
-   struct pipe_transfer base;
-
-   unsigned offset;
-};
-
 
 /*
  * Cast wrappers
@@ -69,11 +59,5 @@ i915_screen(struct pipe_screen *pscreen)
    return (struct i915_screen *) pscreen;
 }
 
-static INLINE struct i915_transfer *
-i915_transfer(struct pipe_transfer *transfer)
-{
-   return (struct i915_transfer *)transfer;
-}
-
 
 #endif /* I915_SCREEN_H */
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index f5b60ed5942..4a1a4a04f6d 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -525,29 +525,40 @@ static void i915_set_constant_buffer(struct pipe_context *pipe,
                                      struct pipe_resource *buf)
 {
    struct i915_context *i915 = i915_context(pipe);
-   draw_flush(i915->draw);
+   unsigned new_num = 0;
+   boolean diff = TRUE;
 
-   /* Make a copy of shader constants.
-    * During fragment program translation we may add additional
-    * constants to the array.
-    *
-    * We want to consider the situation where some user constants
-    * (ex: a material color) may change frequently but the shader program
-    * stays the same.  In that case we should only be updating the first
-    * N constants, leaving any extras from shader translation alone.
-    */
+
+   /* XXX don't support geom shaders now */
+   if (shader == PIPE_SHADER_GEOMETRY)
+      return;
+
+   /* if we have a new buffer compare it with the old one */
    if (buf) {
       struct i915_buffer *ir = i915_buffer(buf);
-      memcpy(i915->current.constants[shader], ir->data, ir->b.b.width0);
-      i915->current.num_user_constants[shader] = (ir->b.b.width0 /
-						  4 * sizeof(float));
-   }
-   else {
-      i915->current.num_user_constants[shader] = 0;
+      struct pipe_resource *old_buf = i915->constants[shader];
+      struct i915_buffer *old = old_buf ? i915_buffer(old_buf) : NULL;
+
+      new_num = ir->b.b.width0 / 4 * sizeof(float);
+
+      if (old && new_num != i915->current.num_user_constants[shader])
+         diff = memcmp(old->data, ir->data, ir->b.b.width0);
+   } else {
+      diff = i915->current.num_user_constants[shader] != 0;
    }
 
+   /*
+    * flush before updateing the state.
+    * XXX: looks like its okay to skip the flush for vertex cbufs
+    */
+   if (diff && shader == PIPE_SHADER_FRAGMENT)
+      draw_flush(i915->draw);
+
+   pipe_resource_reference(&i915->constants[shader], buf);
+   i915->current.num_user_constants[shader] = new_num;
 
-   i915->dirty |= I915_NEW_CONSTANTS;
+   if (diff)
+      i915->dirty |= shader == PIPE_SHADER_VERTEX ? I915_NEW_VS_CONSTANTS : I915_NEW_FS_CONSTANTS;
 }
 
 
diff --git a/src/gallium/drivers/i915/i915_state_emit.c b/src/gallium/drivers/i915/i915_state_emit.c
index 86c02976495..5a89977c26c 100644
--- a/src/gallium/drivers/i915/i915_state_emit.c
+++ b/src/gallium/drivers/i915/i915_state_emit.c
@@ -367,7 +367,8 @@ i915_emit_hardware_state(struct i915_context *i915 )
             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];
+               c = (uint *) i915_buffer(i915->constants[PIPE_SHADER_FRAGMENT])->data;
+               c += 4 * i;
             }
             else {
                /* emit program constant */
diff --git a/src/gallium/drivers/i915/i915_state_fpc.c b/src/gallium/drivers/i915/i915_state_fpc.c
index ec7cec0e471..1959a24691d 100644
--- a/src/gallium/drivers/i915/i915_state_fpc.c
+++ b/src/gallium/drivers/i915/i915_state_fpc.c
@@ -40,7 +40,7 @@ static void update_hw_constants(struct i915_context *i915)
 struct i915_tracked_state i915_hw_constants = {
    "hw_constants",
    update_hw_constants,
-   I915_NEW_CONSTANTS | I915_NEW_FS
+   I915_NEW_FS_CONSTANTS | I915_NEW_FS
 };
 
 
diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 54e339fbefe..935b8454560 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -482,6 +482,7 @@ static int is_alu_trans_unit_inst(struct r600_bc *bc, struct r600_bc_alu *alu)
 		if (!alu->is_op3)
 			return alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_ASHR_INT ||
 				alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT ||
+				alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_FLT_TO_INT_FLOOR ||
 				alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_INT_TO_FLT ||
 				alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHL_INT ||
 				alu->inst == EG_V_SQ_ALU_WORD1_OP2_SQ_OP2_INST_LSHR_INT ||