Merge branch 'gallium-strict-aliasing'

26 files changed, 309 insertions, 216 deletions

diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer.h b/src/gallium/auxiliary/pipebuffer/pb_buffer.h
index 4ef372233f0..eb7e84be848 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer.h
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer.h
@@ -237,8 +237,9 @@ pb_reference(struct pb_buffer **dst,
 {
    struct pb_buffer *old = *dst;
 
-   if (pipe_reference((struct pipe_reference**)dst, &src->base.reference))
+   if (pipe_reference(&(*dst)->base.reference, &src->base.reference))
       pb_destroy( old );
+   *dst = src;
 }
 
 
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
index 2f973684f67..a9375abd218 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
@@ -243,6 +243,7 @@ fenced_buffer_list_check_free_locked(struct fenced_buffer_list *fenced_list,
    struct pb_fence_ops *ops = fenced_list->ops;
    struct list_head *curr, *next;
    struct fenced_buffer *fenced_buf;
+   struct pb_buffer *pb_buf;
    struct pipe_fence_handle *prev_fence = NULL;
 
    curr = fenced_list->delayed.next;
@@ -271,7 +272,9 @@ fenced_buffer_list_check_free_locked(struct fenced_buffer_list *fenced_list,
       fenced_buffer_remove_locked(fenced_list, fenced_buf);
       pipe_mutex_unlock(fenced_buf->mutex);
 
-      pb_reference((struct pb_buffer **)&fenced_buf, NULL);
+      pb_buf = &fenced_buf->base;
+      pb_reference(&pb_buf, NULL);
+      
 
       curr = next; 
       next = curr->next;
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
index 57d1ede45a4..7b34c8e3578 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
@@ -294,7 +294,7 @@ pb_cache_manager_create_buffer(struct pb_manager *_mgr,
       LIST_DEL(&buf->head);
       pipe_mutex_unlock(mgr->mutex);
       /* Increase refcount */
-      pb_reference((struct pb_buffer**)&buf, &buf->base);
+      pipe_reference(NULL, &buf->base.base.reference);
       return &buf->base;
    }
    
diff --git a/src/gallium/auxiliary/util/u_clear.h b/src/gallium/auxiliary/util/u_clear.h
index 1e65a035aed..2c32db61756 100644
--- a/src/gallium/auxiliary/util/u_clear.h
+++ b/src/gallium/auxiliary/util/u_clear.h
@@ -46,13 +46,13 @@ util_clear(struct pipe_context *pipe,
 {
    if (buffers & PIPE_CLEAR_COLOR) {
       struct pipe_surface *ps = framebuffer->cbufs[0];
-      unsigned color;
+      union util_color uc;
 
-      util_pack_color(rgba, ps->format, &color);
+      util_pack_color(rgba, ps->format, &uc);
       if (pipe->surface_fill) {
-         pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, color);
+         pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, uc.ui);
       } else {
-         util_surface_fill(pipe, ps, 0, 0, ps->width, ps->height, color);
+         util_surface_fill(pipe, ps, 0, 0, ps->width, ps->height, uc.ui);
       }
    }
 
diff --git a/src/gallium/auxiliary/util/u_pack_color.h b/src/gallium/auxiliary/util/u_pack_color.h
index 9dacc6d83db..a2e0f266864 100644
--- a/src/gallium/auxiliary/util/u_pack_color.h
+++ b/src/gallium/auxiliary/util/u_pack_color.h
@@ -40,101 +40,97 @@
 #include "util/u_math.h"
 
 
+
+union util_color {
+   ubyte ub;
+   ushort us;
+   uint ui;
+   float f[4];
+};
+
 /**
  * Pack ubyte R,G,B,A into dest pixel.
  */
 static INLINE void
 util_pack_color_ub(ubyte r, ubyte g, ubyte b, ubyte a,
-                   enum pipe_format format, void *dest)
+                   enum pipe_format format, union util_color *uc)
 {
    switch (format) {
    case PIPE_FORMAT_R8G8B8A8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (r << 24) | (g << 16) | (b << 8) | a;
+         uc->ui = (r << 24) | (g << 16) | (b << 8) | a;
       }
       return;
    case PIPE_FORMAT_R8G8B8X8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (r << 24) | (g << 16) | (b << 8) | 0xff;
+         uc->ui = (r << 24) | (g << 16) | (b << 8) | 0xff;
       }
       return;
    case PIPE_FORMAT_A8R8G8B8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (a << 24) | (r << 16) | (g << 8) | b;
+         uc->ui = (a << 24) | (r << 16) | (g << 8) | b;
       }
       return;
    case PIPE_FORMAT_X8R8G8B8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (0xff << 24) | (r << 16) | (g << 8) | b;
+         uc->ui = (0xff << 24) | (r << 16) | (g << 8) | b;
       }
       return;
    case PIPE_FORMAT_B8G8R8A8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (b << 24) | (g << 16) | (r << 8) | a;
+         uc->ui = (b << 24) | (g << 16) | (r << 8) | a;
       }
       return;
    case PIPE_FORMAT_B8G8R8X8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (b << 24) | (g << 16) | (r << 8) | 0xff;
+         uc->ui = (b << 24) | (g << 16) | (r << 8) | 0xff;
       }
       return;
    case PIPE_FORMAT_R5G6B5_UNORM:
       {
-         ushort *d = (ushort *) dest;
-         *d = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
+         uc->us = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
       }
       return;
    case PIPE_FORMAT_A1R5G5B5_UNORM:
       {
-         ushort *d = (ushort *) dest;
-         *d = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3);
+         uc->us = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3);
       }
       return;
    case PIPE_FORMAT_A4R4G4B4_UNORM:
       {
-         ushort *d = (ushort *) dest;
-         *d = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4);
+         uc->us = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4);
       }
       return;
    case PIPE_FORMAT_A8_UNORM:
       {
-         ubyte *d = (ubyte *) dest;
-         *d = a;
+         uc->ub = a;
       }
       return;
    case PIPE_FORMAT_L8_UNORM:
    case PIPE_FORMAT_I8_UNORM:
       {
-         ubyte *d = (ubyte *) dest;
-         *d = r;
+         uc->ub = a;
       }
       return;
    case PIPE_FORMAT_R32G32B32A32_FLOAT:
       {
-         float *d = (float *) dest;
-         d[0] = (float)r / 255.0f;
-         d[1] = (float)g / 255.0f;
-         d[2] = (float)b / 255.0f;
-         d[3] = (float)a / 255.0f;
+         uc->f[0] = (float)r / 255.0f;
+         uc->f[1] = (float)g / 255.0f;
+         uc->f[2] = (float)b / 255.0f;
+         uc->f[3] = (float)a / 255.0f;
       }
       return;
    case PIPE_FORMAT_R32G32B32_FLOAT:
       {
-         float *d = (float *) dest;
-         d[0] = (float)r / 255.0f;
-         d[1] = (float)g / 255.0f;
-         d[2] = (float)b / 255.0f;
+         uc->f[0] = (float)r / 255.0f;
+         uc->f[1] = (float)g / 255.0f;
+         uc->f[2] = (float)b / 255.0f;
       }
       return;
 
    /* XXX lots more cases to add */
    default:
+      uc->ui = 0; /* keep compiler happy */
       debug_print_format("gallium: unhandled format in util_pack_color_ub()", format);
       assert(0);
    }
@@ -145,13 +141,13 @@ util_pack_color_ub(ubyte r, ubyte g, ubyte b, ubyte a,
  * Unpack RGBA from a packed pixel, returning values as ubytes in [0,255].
  */
 static INLINE void
-util_unpack_color_ub(enum pipe_format format, const void *src,
+util_unpack_color_ub(enum pipe_format format, union util_color *uc,
                      ubyte *r, ubyte *g, ubyte *b, ubyte *a)
 {
    switch (format) {
    case PIPE_FORMAT_R8G8B8A8_UNORM:
       {
-         uint p = ((const uint *) src)[0];
+         uint p = uc->ui;
          *r = (ubyte) ((p >> 24) & 0xff);
          *g = (ubyte) ((p >> 16) & 0xff);
          *b = (ubyte) ((p >>  8) & 0xff);
@@ -160,7 +156,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_R8G8B8X8_UNORM:
       {
-         uint p = ((const uint *) src)[0];
+         uint p = uc->ui;
          *r = (ubyte) ((p >> 24) & 0xff);
          *g = (ubyte) ((p >> 16) & 0xff);
          *b = (ubyte) ((p >>  8) & 0xff);
@@ -169,7 +165,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_A8R8G8B8_UNORM:
       {
-         uint p = ((const uint *) src)[0];
+         uint p = uc->ui;
          *r = (ubyte) ((p >> 16) & 0xff);
          *g = (ubyte) ((p >>  8) & 0xff);
          *b = (ubyte) ((p >>  0) & 0xff);
@@ -178,7 +174,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_X8R8G8B8_UNORM:
       {
-         uint p = ((const uint *) src)[0];
+         uint p = uc->ui;
          *r = (ubyte) ((p >> 16) & 0xff);
          *g = (ubyte) ((p >>  8) & 0xff);
          *b = (ubyte) ((p >>  0) & 0xff);
@@ -187,7 +183,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_B8G8R8A8_UNORM:
       {
-         uint p = ((const uint *) src)[0];
+         uint p = uc->ui;
          *r = (ubyte) ((p >>  8) & 0xff);
          *g = (ubyte) ((p >> 16) & 0xff);
          *b = (ubyte) ((p >> 24) & 0xff);
@@ -196,7 +192,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_B8G8R8X8_UNORM:
       {
-         uint p = ((const uint *) src)[0];
+         uint p = uc->ui;
          *r = (ubyte) ((p >>  8) & 0xff);
          *g = (ubyte) ((p >> 16) & 0xff);
          *b = (ubyte) ((p >> 24) & 0xff);
@@ -205,7 +201,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_R5G6B5_UNORM:
       {
-         ushort p = ((const ushort *) src)[0];
+         ushort p = uc->us;
          *r = (ubyte) (((p >> 8) & 0xf8) | ((p >> 13) & 0x7));
          *g = (ubyte) (((p >> 3) & 0xfc) | ((p >>  9) & 0x3));
          *b = (ubyte) (((p << 3) & 0xf8) | ((p >>  2) & 0x7));
@@ -214,7 +210,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_A1R5G5B5_UNORM:
       {
-         ushort p = ((const ushort *) src)[0];
+         ushort p = uc->us;
          *r = (ubyte) (((p >>  7) & 0xf8) | ((p >> 12) & 0x7));
          *g = (ubyte) (((p >>  2) & 0xf8) | ((p >>  7) & 0x7));
          *b = (ubyte) (((p <<  3) & 0xf8) | ((p >>  2) & 0x7));
@@ -223,7 +219,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_A4R4G4B4_UNORM:
       {
-         ushort p = ((const ushort *) src)[0];
+         ushort p = uc->us;
          *r = (ubyte) (((p >> 4) & 0xf0) | ((p >>  8) & 0xf));
          *g = (ubyte) (((p >> 0) & 0xf0) | ((p >>  4) & 0xf));
          *b = (ubyte) (((p << 4) & 0xf0) | ((p >>  0) & 0xf));
@@ -232,27 +228,27 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_A8_UNORM:
       {
-         ubyte p = ((const ubyte *) src)[0];
+         ubyte p = uc->ub;
          *r = *g = *b = (ubyte) 0xff;
          *a = p;
       }
       return;
    case PIPE_FORMAT_L8_UNORM:
       {
-         ubyte p = ((const ubyte *) src)[0];
+         ubyte p = uc->ub;
          *r = *g = *b = p;
          *a = (ubyte) 0xff;
       }
       return;
    case PIPE_FORMAT_I8_UNORM:
       {
-         ubyte p = ((const ubyte *) src)[0];
+         ubyte p = uc->ub;
          *r = *g = *b = *a = p;
       }
       return;
    case PIPE_FORMAT_R32G32B32A32_FLOAT:
       {
-         const float *p = (const float *) src;
+         const float *p = &uc->f[0];
          *r = float_to_ubyte(p[0]);
          *g = float_to_ubyte(p[1]);
          *b = float_to_ubyte(p[2]);
@@ -261,7 +257,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       return;
    case PIPE_FORMAT_R32G32B32_FLOAT:
       {
-         const float *p = (const float *) src;
+         const float *p = &uc->f[0];
          *r = float_to_ubyte(p[0]);
          *g = float_to_ubyte(p[1]);
          *b = float_to_ubyte(p[2]);
@@ -271,7 +267,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
 
    case PIPE_FORMAT_R32G32_FLOAT:
       {
-         const float *p = (const float *) src;
+         const float *p = &uc->f[0];
          *r = float_to_ubyte(p[0]);
          *g = float_to_ubyte(p[1]);
          *b = *a = (ubyte) 0xff;
@@ -280,7 +276,7 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
 
    case PIPE_FORMAT_R32_FLOAT:
       {
-         const float *p = (const float *) src;
+         const float *p = &uc->f[0];
          *r = float_to_ubyte(p[0]);
          *g = *b = *a = (ubyte) 0xff;
       }
@@ -293,14 +289,13 @@ util_unpack_color_ub(enum pipe_format format, const void *src,
       assert(0);
    }
 }
- 
 
 
 /**
  * Note rgba outside [0,1] will be clamped for int pixel formats.
  */
 static INLINE void
-util_pack_color(const float rgba[4], enum pipe_format format, void *dest)
+util_pack_color(const float rgba[4], enum pipe_format format, union util_color *uc)
 {
    ubyte r = 0;
    ubyte g = 0;
@@ -318,90 +313,78 @@ util_pack_color(const float rgba[4], enum pipe_format format, void *dest)
    switch (format) {
    case PIPE_FORMAT_R8G8B8A8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (r << 24) | (g << 16) | (b << 8) | a;
+         uc->ui = (r << 24) | (g << 16) | (b << 8) | a;
       }
       return;
    case PIPE_FORMAT_R8G8B8X8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (r << 24) | (g << 16) | (b << 8) | 0xff;
+         uc->ui = (r << 24) | (g << 16) | (b << 8) | 0xff;
       }
       return;
    case PIPE_FORMAT_A8R8G8B8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (a << 24) | (r << 16) | (g << 8) | b;
+         uc->ui = (a << 24) | (r << 16) | (g << 8) | b;
       }
       return;
    case PIPE_FORMAT_X8R8G8B8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (0xff << 24) | (r << 16) | (g << 8) | b;
+         uc->ui = (0xff << 24) | (r << 16) | (g << 8) | b;
       }
       return;
    case PIPE_FORMAT_B8G8R8A8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (b << 24) | (g << 16) | (r << 8) | a;
+         uc->ui = (b << 24) | (g << 16) | (r << 8) | a;
       }
       return;
    case PIPE_FORMAT_B8G8R8X8_UNORM:
       {
-         uint *d = (uint *) dest;
-         *d = (b << 24) | (g << 16) | (r << 8) | 0xff;
+         uc->ui = (b << 24) | (g << 16) | (r << 8) | 0xff;
       }
       return;
    case PIPE_FORMAT_R5G6B5_UNORM:
       {
-         ushort *d = (ushort *) dest;
-         *d = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
+         uc->us = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
       }
       return;
    case PIPE_FORMAT_A1R5G5B5_UNORM:
       {
-         ushort *d = (ushort *) dest;
-         *d = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3);
+         uc->us = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3);
       }
       return;
    case PIPE_FORMAT_A4R4G4B4_UNORM:
       {
-         ushort *d = (ushort *) dest;
-         *d = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4);
+         uc->ub = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4);
       }
       return;
    case PIPE_FORMAT_A8_UNORM:
       {
-         ubyte *d = (ubyte *) dest;
-         *d = a;
+         uc->ub = a;
       }
       return;
    case PIPE_FORMAT_L8_UNORM:
    case PIPE_FORMAT_I8_UNORM:
       {
-         ubyte *d = (ubyte *) dest;
-         *d = r;
+         uc->ub = r;
       }
       return;
    case PIPE_FORMAT_R32G32B32A32_FLOAT:
       {
-         float *d = (float *) dest;
-         d[0] = rgba[0];
-         d[1] = rgba[1];
-         d[2] = rgba[2];
-         d[3] = rgba[3];
+         uc->f[0] = rgba[0];
+         uc->f[1] = rgba[1];
+         uc->f[2] = rgba[2];
+         uc->f[3] = rgba[3];
       }
       return;
    case PIPE_FORMAT_R32G32B32_FLOAT:
       {
-         float *d = (float *) dest;
-         d[0] = rgba[0];
-         d[1] = rgba[1];
-         d[2] = rgba[2];
+         uc->f[0] = rgba[0];
+         uc->f[1] = rgba[1];
+         uc->f[2] = rgba[2];
       }
       return;
    /* XXX lots more cases to add */
    default:
+      uc->ui = 0; /* keep compiler happy */
       debug_print_format("gallium: unhandled format in util_pack_color()", format);
       assert(0);
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_clear.c b/src/gallium/drivers/cell/ppu/cell_clear.c
index 79ad687ea94..3a3f968a492 100644
--- a/src/gallium/drivers/cell/ppu/cell_clear.c
+++ b/src/gallium/drivers/cell/ppu/cell_clear.c
@@ -59,9 +59,9 @@ cell_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
 
    if (buffers & PIPE_CLEAR_COLOR) {
       uint surfIndex = 0;
-      uint clearValue;
+      union util_color uc;
 
-      util_pack_color(rgba, cell->framebuffer.cbufs[0]->format, &clearValue);
+      util_pack_color(rgba, cell->framebuffer.cbufs[0]->format, &uc);
 
       /* Build a CLEAR command and place it in the current batch buffer */
       STATIC_ASSERT(sizeof(struct cell_command_clear_surface) % 16 == 0);
@@ -70,7 +70,7 @@ cell_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
          cell_batch_alloc16(cell, sizeof(*clr));
       clr->opcode[0] = CELL_CMD_CLEAR_SURFACE;
       clr->surface = surfIndex;
-      clr->value = clearValue;
+      clr->value = uc.ui;
    }
 
    if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c
index 4509c7b1e5c..bedab56f59e 100644
--- a/src/gallium/drivers/identity/id_context.c
+++ b/src/gallium/drivers/identity/id_context.c
@@ -742,7 +742,7 @@ identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    id_pipe->base.set_polygon_stipple = identity_set_polygon_stipple;
    id_pipe->base.set_scissor_state = identity_set_scissor_state;
    id_pipe->base.set_viewport_state = identity_set_viewport_state;
-   id_pipe->base.set_fragment_sampler_textures = identity_set_vertex_sampler_textures;
+   id_pipe->base.set_fragment_sampler_textures = identity_set_fragment_sampler_textures;
    id_pipe->base.set_vertex_sampler_textures = identity_set_vertex_sampler_textures;
    id_pipe->base.set_vertex_buffers = identity_set_vertex_buffers;
    id_pipe->base.set_vertex_elements = identity_set_vertex_elements;
diff --git a/src/gallium/drivers/llvmpipe/lp_clear.c b/src/gallium/drivers/llvmpipe/lp_clear.c
index bdcff94b9bf..08d9f2e2735 100644
--- a/src/gallium/drivers/llvmpipe/lp_clear.c
+++ b/src/gallium/drivers/llvmpipe/lp_clear.c
@@ -50,6 +50,7 @@ llvmpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
                double depth, unsigned stencil)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
+   union util_color uc;
    unsigned cv;
    uint i;
 
@@ -64,8 +65,8 @@ llvmpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
       for (i = 0; i < llvmpipe->framebuffer.nr_cbufs; i++) {
          struct pipe_surface *ps = llvmpipe->framebuffer.cbufs[i];
 
-         util_pack_color(rgba, ps->format, &cv);
-         lp_tile_cache_clear(llvmpipe->cbuf_cache[i], rgba, cv);
+         util_pack_color(rgba, ps->format, &uc);
+         lp_tile_cache_clear(llvmpipe->cbuf_cache[i], rgba, uc.ui);
       }
       llvmpipe->dirty_render_cache = TRUE;
    }
diff --git a/src/gallium/drivers/nouveau/nouveau_stateobj.h b/src/gallium/drivers/nouveau/nouveau_stateobj.h
index b595405357f..62990f9b6a6 100644
--- a/src/gallium/drivers/nouveau/nouveau_stateobj.h
+++ b/src/gallium/drivers/nouveau/nouveau_stateobj.h
@@ -48,13 +48,14 @@ so_ref(struct nouveau_stateobj *ref, struct nouveau_stateobj **pso)
 	struct nouveau_stateobj *so = *pso;
 	int i;
 
-        if (pipe_reference((struct pipe_reference**)pso, &ref->reference)) {
+        if (pipe_reference(&(*pso)->reference, &ref->reference)) {
 		free(so->push);
 		for (i = 0; i < so->cur_reloc; i++)
 			nouveau_bo_ref(NULL, &so->reloc[i].bo);
 		free(so->reloc);
 		free(so);
 	}
+	*pso = ref;
 }
 
 static INLINE void
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 8ef0b3b268a..68c5408a647 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -151,9 +151,10 @@ static void r300_set_blend_color(struct pipe_context* pipe,
                                  const struct pipe_blend_color* color)
 {
     struct r300_context* r300 = r300_context(pipe);
+    union util_color uc;
 
-    util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM,
-            &r300->blend_color_state->blend_color);
+    util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
+    r300->blend_color_state->blend_color = uc.ui;
 
     /* XXX if FP16 blending is enabled, we should use the FP16 format */
     r300->blend_color_state->blend_color_red_alpha =
@@ -549,6 +550,7 @@ static void*
     struct r300_context* r300 = r300_context(pipe);
     struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
     int lod_bias;
+    union util_color uc;
 
     sampler->filter0 |=
         (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
@@ -570,8 +572,8 @@ static void*
 
     sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
 
-    util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM,
-                    &sampler->border_color);
+    util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
+    sampler->border_color = uc.ui;
 
     /* R500-specific fixups and optimizations */
     if (r300_screen(r300->context.screen)->caps->is_r500) {
diff --git a/src/gallium/drivers/softpipe/sp_clear.c b/src/gallium/drivers/softpipe/sp_clear.c
index 8fac8e6e05f..f98087deb8c 100644
--- a/src/gallium/drivers/softpipe/sp_clear.c
+++ b/src/gallium/drivers/softpipe/sp_clear.c
@@ -48,6 +48,7 @@ softpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
                double depth, unsigned stencil)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
+   union util_color uc;
    unsigned cv;
    uint i;
 
@@ -62,12 +63,12 @@ softpipe_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
       for (i = 0; i < softpipe->framebuffer.nr_cbufs; i++) {
          struct pipe_surface *ps = softpipe->framebuffer.cbufs[i];
 
-         util_pack_color(rgba, ps->format, &cv);
-         sp_tile_cache_clear(softpipe->cbuf_cache[i], rgba, cv);
+         util_pack_color(rgba, ps->format, &uc);
+         sp_tile_cache_clear(softpipe->cbuf_cache[i], rgba, uc.ui);
 
 #if !TILE_CLEAR_OPTIMIZATION
          /* non-cached surface */
-         pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, cv);
+         pipe->surface_fill(pipe, ps, 0, 0, ps->width, ps->height, uc.ui);
 #endif
       }
    }
diff --git a/src/gallium/drivers/svga/svga_pipe_clear.c b/src/gallium/drivers/svga/svga_pipe_clear.c
index 6195c3897ed..409b3b41cbc 100644
--- a/src/gallium/drivers/svga/svga_pipe_clear.c
+++ b/src/gallium/drivers/svga/svga_pipe_clear.c
@@ -46,7 +46,7 @@ try_clear(struct svga_context *svga,
    boolean restore_viewport = FALSE;
    SVGA3dClearFlag flags = 0;
    struct pipe_framebuffer_state *fb = &svga->curr.framebuffer;
-   unsigned color = 0;
+   union util_color uc;
 
    ret = svga_update_state(svga, SVGA_STATE_HW_CLEAR);
    if (ret)
@@ -54,7 +54,7 @@ try_clear(struct svga_context *svga,
 
    if ((buffers & PIPE_CLEAR_COLOR) && fb->cbufs[0]) {
       flags |= SVGA3D_CLEAR_COLOR;
-      util_pack_color(rgba, PIPE_FORMAT_A8R8G8B8_UNORM, &color);
+      util_pack_color(rgba, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
 
       rect.w = fb->cbufs[0]->width;
       rect.h = fb->cbufs[0]->height;
@@ -77,7 +77,7 @@ try_clear(struct svga_context *svga,
          return ret;
    }
 
-   ret = SVGA3D_ClearRect(svga->swc, flags, color, depth, stencil,
+   ret = SVGA3D_ClearRect(svga->swc, flags, uc.ui, depth, stencil,
                           rect.x, rect.y, rect.w, rect.h);
    if (ret != PIPE_OK)
       return ret;
diff --git a/src/gallium/drivers/svga/svga_pipe_sampler.c b/src/gallium/drivers/svga/svga_pipe_sampler.c
index b4e57c5d15b..78053e755e2 100644
--- a/src/gallium/drivers/svga/svga_pipe_sampler.c
+++ b/src/gallium/drivers/svga/svga_pipe_sampler.c
@@ -101,6 +101,7 @@ svga_create_sampler_state(struct pipe_context *pipe,
 {
    struct svga_context *svga = svga_context(pipe);
    struct svga_sampler_state *cso = CALLOC_STRUCT( svga_sampler_state );
+   union util_color uc;
    
    cso->mipfilter = translate_mip_filter(sampler->min_mip_filter);
    cso->magfilter = translate_img_filter( sampler->mag_img_filter );
@@ -121,8 +122,8 @@ svga_create_sampler_state(struct pipe_context *pipe,
       ubyte a = float_to_ubyte(sampler->border_color[3]);
 
       util_pack_color_ub( r, g, b, a,
-                          PIPE_FORMAT_B8G8R8A8_UNORM,
-                          &cso->bordercolor );
+                          PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
+      cso->bordercolor = uc.ui;
    }
 
    /* No SVGA3D support for:
diff --git a/src/gallium/drivers/svga/svga_screen_buffer.c b/src/gallium/drivers/svga/svga_screen_buffer.c
index 1f8a8896723..58a1aba464b 100644
--- a/src/gallium/drivers/svga/svga_screen_buffer.c
+++ b/src/gallium/drivers/svga/svga_screen_buffer.c
@@ -356,7 +356,8 @@ svga_buffer_upload_flush(struct svga_context *svga,
    sbuf->hw.boxes = NULL;
 
    /* Decrement reference count */
-   pipe_buffer_reference((struct pipe_buffer **)&sbuf, NULL);
+   pipe_reference(&(sbuf->base.reference), NULL);
+   sbuf = NULL;
 }
 
 
diff --git a/src/gallium/drivers/svga/svga_screen_texture.h b/src/gallium/drivers/svga/svga_screen_texture.h
index 1cc4063e653..727f2c51d28 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.h
+++ b/src/gallium/drivers/svga/svga_screen_texture.h
@@ -164,8 +164,9 @@ svga_sampler_view_reference(struct svga_sampler_view **ptr, struct svga_sampler_
 {
    struct svga_sampler_view *old = *ptr;
 
-   if (pipe_reference((struct pipe_reference **)ptr, &v->reference))
+   if (pipe_reference(&(*ptr)->reference, &v->reference))
       svga_destroy_sampler_view_priv(old);
+   *ptr = v;
 }
 
 extern void
diff --git a/src/gallium/include/pipe/p_refcnt.h b/src/gallium/include/pipe/p_refcnt.h
index 1f9088b3e9c..f1875b6b822 100644
--- a/src/gallium/include/pipe/p_refcnt.h
+++ b/src/gallium/include/pipe/p_refcnt.h
@@ -59,30 +59,28 @@ pipe_is_referenced(struct pipe_reference *reference)
 
 
 /**
- * Set 'ptr' to point to 'reference' and update reference counting.
- * The old thing pointed to, if any, will be unreferenced first.
- * 'reference' may be NULL.
+ * Update reference counting.
+ * The old thing pointed to, if any, will be unreferenced.
+ * Both 'ptr' and 'reference' may be NULL.
  */
 static INLINE bool
-pipe_reference(struct pipe_reference **ptr, struct pipe_reference *reference)
+pipe_reference(struct pipe_reference *ptr, struct pipe_reference *reference)
 {
    bool destroy = FALSE;
 
-   if(*ptr != reference) {
+   if(ptr != reference) {
       /* bump the reference.count first */
       if (reference) {
          assert(pipe_is_referenced(reference));
          p_atomic_inc(&reference->count);
       }
    
-      if (*ptr) {
-         assert(pipe_is_referenced(*ptr));
-         if (p_atomic_dec_zero(&(*ptr)->count)) {
+      if (ptr) {
+         assert(pipe_is_referenced(ptr));
+         if (p_atomic_dec_zero(&ptr->count)) {
             destroy = TRUE;
          }
       }
-   
-      *ptr = reference;
    }
 
    return destroy;
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index cbd5bfa4699..60e96b98deb 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -392,8 +392,9 @@ pipe_buffer_reference(struct pipe_buffer **ptr, struct pipe_buffer *buf)
 {
    struct pipe_buffer *old_buf = *ptr;
 
-   if (pipe_reference((struct pipe_reference **)ptr, &buf->reference))
+   if (pipe_reference(&(*ptr)->reference, &buf->reference))
       old_buf->screen->buffer_destroy(old_buf);
+   *ptr = buf;
 }
 
 static INLINE void
@@ -401,8 +402,9 @@ pipe_surface_reference(struct pipe_surface **ptr, struct pipe_surface *surf)
 {
    struct pipe_surface *old_surf = *ptr;
 
-   if (pipe_reference((struct pipe_reference **)ptr, &surf->reference))
+   if (pipe_reference(&(*ptr)->reference, &surf->reference))
       old_surf->texture->screen->tex_surface_destroy(old_surf);
+   *ptr = surf;
 }
 
 static INLINE void
@@ -410,8 +412,9 @@ pipe_texture_reference(struct pipe_texture **ptr, struct pipe_texture *tex)
 {
    struct pipe_texture *old_tex = *ptr;
 
-   if (pipe_reference((struct pipe_reference **)ptr, &tex->reference))
+   if (pipe_reference(&(*ptr)->reference, &tex->reference))
       old_tex->screen->texture_destroy(old_tex);
+   *ptr = tex;
 }
 
 
diff --git a/src/gallium/include/pipe/p_video_state.h b/src/gallium/include/pipe/p_video_state.h
index 4da26d608cf..b85f01c2b02 100644
--- a/src/gallium/include/pipe/p_video_state.h
+++ b/src/gallium/include/pipe/p_video_state.h
@@ -56,8 +56,9 @@ pipe_video_surface_reference(struct pipe_video_surface **ptr, struct pipe_video_
 {
    struct pipe_video_surface *old_surf = *ptr;
 
-   if (pipe_reference((struct pipe_reference **)ptr, &surf->reference))
+   if (pipe_reference(&(*ptr)->reference, &surf->reference))
       old_surf->screen->video_surface_destroy(old_surf);
+   *ptr = surf;
 }
 
 struct pipe_video_rect
diff --git a/src/gallium/state_trackers/python/st_device.c b/src/gallium/state_trackers/python/st_device.c
index 2966b24cdcc..10c7ecbd78f 100644
--- a/src/gallium/state_trackers/python/st_device.c
+++ b/src/gallium/state_trackers/python/st_device.c
@@ -62,8 +62,9 @@ st_device_reference(struct st_device **ptr, struct st_device *st_dev)
 {
    struct st_device *old_dev = *ptr;
 
-   if (pipe_reference((struct pipe_reference **)ptr, &st_dev->reference))
+   if (pipe_reference(&(*ptr)->reference, &st_dev->reference))
       st_device_really_destroy(old_dev);
+   *ptr = st_dev;
 }
 
 
diff --git a/src/gallium/state_trackers/vega/vg_translate.c b/src/gallium/state_trackers/vega/vg_translate.c
index 00e07647062..03575ca3ddd 100644
--- a/src/gallium/state_trackers/vega/vg_translate.c
+++ b/src/gallium/state_trackers/vega/vg_translate.c
@@ -474,6 +474,7 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
                                   VGfloat rgba[][4])
 {
    VGint i;
+   union util_color uc;
 
    switch (dataFormat) {
    case VG_sRGBX_8888: {
@@ -486,8 +487,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          b = (*src >>  8) & 0xff;
          a = 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -502,8 +506,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          b = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -519,8 +526,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          b = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -536,8 +546,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          clr[2] = ((*src >>  0) & 31)/31.;
          clr[3] = 1.f;
 
-         util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                         rgba[i]);
+         util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -552,8 +565,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          clr[2] = ((*src >>  1) & 31)/31.;
          clr[3] = ((*src >>  0) & 1)/1.;
 
-         util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                         rgba[i]);
+         util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -568,8 +584,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          clr[2] = ((*src >>  4) & 15)/15.;
          clr[3] = ((*src >>  0) & 15)/15.;
 
-         util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                         rgba[i]);
+         util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -578,8 +597,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
       VGubyte *src = (VGubyte *)data;
       src += offset;
       for (i = 0; i < n; ++i) {
-         util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                         rgba[i]);
+         util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -594,8 +616,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          b = (*src >>  8) & 0xff;
          a = 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -610,8 +635,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          b = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -627,8 +655,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          b = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -638,8 +669,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
       VGubyte *src = (VGubyte *)data;
       src += offset;
       for (i = 0; i < n; ++i) {
-         util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                         rgba[i]);
+         util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -648,8 +682,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
       VGubyte *src = (VGubyte *)data;
       src += offset;
       for (i = 0; i < n; ++i) {
-         util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(0xff, 0xff, 0xff, *src, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
    }
@@ -667,8 +704,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
             clr[2] = clr[0];
             clr[3] = 1.f;
 
-            util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i+j]);
+            util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+            rgba[i+j][0] = uc.f[0];
+            rgba[i+j][1] = uc.f[1];
+            rgba[i+j][2] = uc.f[2];
+            rgba[i+j][3] = uc.f[3];
          }
          ++src;
       }
@@ -688,8 +728,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
             clr[2] = 0.f;
             clr[3] = (((*src) & (1<<shift)) >> shift);
 
-            util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i+j]);
+            util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+            rgba[i+j][0] = uc.f[0];
+            rgba[i+j][1] = uc.f[1];
+            rgba[i+j][2] = uc.f[2];
+            rgba[i+j][3] = uc.f[3];
          }
          ++src;
       }
@@ -715,8 +758,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
             clr[2] = 0.f;
             clr[3] = ((*src) & (bitter)) >> shift;
 
-            util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i +j]);
+            util_pack_color(clr, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+            rgba[i+j][0] = uc.f[0];
+            rgba[i+j][1] = uc.f[1];
+            rgba[i+j][2] = uc.f[2];
+            rgba[i+j][3] = uc.f[3];
          }
          ++src;
       }
@@ -735,8 +781,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          b = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -752,8 +801,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          b = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -775,8 +827,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          b = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -792,8 +847,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          b = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -811,8 +869,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          r = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -828,8 +889,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          r = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -853,8 +917,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          r = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -870,8 +937,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          r = (*src >>  8) & 0xff;
          a = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -889,8 +959,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          r = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -906,8 +979,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          r = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -929,8 +1005,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          r = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
@@ -946,8 +1025,11 @@ void _vega_unpack_float_span_rgba(struct vg_context *ctx,
          g = (*src >>  8) & 0xff;
          r = (*src >>  0) & 0xff;
 
-         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT,
-                            rgba[i]);
+         util_pack_color_ub(r, g, b, a, PIPE_FORMAT_R32G32B32A32_FLOAT, &uc);
+         rgba[i][0] = uc.f[0];
+         rgba[i][1] = uc.f[1];
+         rgba[i][2] = uc.f[2];
+         rgba[i][3] = uc.f[3];
          ++src;
       }
       return;
diff --git a/src/gallium/winsys/drm/intel/gem/intel_drm_fence.c b/src/gallium/winsys/drm/intel/gem/intel_drm_fence.c
index e70bfe7b44e..b6248a3bcf7 100644
--- a/src/gallium/winsys/drm/intel/gem/intel_drm_fence.c
+++ b/src/gallium/winsys/drm/intel/gem/intel_drm_fence.c
@@ -39,11 +39,12 @@ intel_drm_fence_reference(struct intel_winsys *iws,
    struct intel_drm_fence *old = (struct intel_drm_fence *)*ptr;
    struct intel_drm_fence *f = (struct intel_drm_fence *)fence;
 
-   if (pipe_reference((struct pipe_reference**)ptr, &f->reference)) {
+   if (pipe_reference(&(*ptr)->reference, &f->reference)) {
       if (old->bo)
          drm_intel_bo_unreference(old->bo);
       FREE(old);
    }
+   *ptr = fence;
 }
 
 static int
diff --git a/src/gallium/winsys/drm/vmware/core/vmw_surface.c b/src/gallium/winsys/drm/vmware/core/vmw_surface.c
index 64eb32f8b94..5f1b9ad5770 100644
--- a/src/gallium/winsys/drm/vmware/core/vmw_surface.c
+++ b/src/gallium/winsys/drm/vmware/core/vmw_surface.c
@@ -47,7 +47,7 @@ vmw_svga_winsys_surface_reference(struct vmw_svga_winsys_surface **pdst,
    src_ref = src ? &src->refcnt : NULL;
    dst_ref = dst ? &dst->refcnt : NULL;
 
-   if (pipe_reference(&dst_ref, src_ref)) {
+   if (pipe_reference(dst_ref, src_ref)) {
       vmw_ioctl_surface_destroy(dst->screen, dst->sid);
 #ifdef DEBUG
       /* to detect dangling pointers */
diff --git a/src/mesa/main/imports.c b/src/mesa/main/imports.c
index c9e00cf7528..f2ef84f35c1 100644
--- a/src/mesa/main/imports.c
+++ b/src/mesa/main/imports.c
@@ -460,7 +460,7 @@ _mesa_inv_sqrtf(float n)
 #if 0 /* not used, see below -BP */
         float r3, x3, y3;
 #endif
-        union { float f; unsigned int i; } u;
+        fi_type u;
         unsigned int magic;
 
         /*
@@ -649,10 +649,10 @@ _mesa_bitcount(unsigned int n)
 GLhalfARB
 _mesa_float_to_half(float val)
 {
-   const int flt = *((int *) (void *) &val);
-   const int flt_m = flt & 0x7fffff;
-   const int flt_e = (flt >> 23) & 0xff;
-   const int flt_s = (flt >> 31) & 0x1;
+   const fi_type fi = {val};
+   const int flt_m = fi.i & 0x7fffff;
+   const int flt_e = (fi.i >> 23) & 0xff;
+   const int flt_s = (fi.i >> 31) & 0x1;
    int s, e, m = 0;
    GLhalfARB result;
    
@@ -739,7 +739,8 @@ _mesa_half_to_float(GLhalfARB val)
    const int m = val & 0x3ff;
    const int e = (val >> 10) & 0x1f;
    const int s = (val >> 15) & 0x1;
-   int flt_m, flt_e, flt_s, flt;
+   int flt_m, flt_e, flt_s;
+   fi_type fi;
    float result;
 
    /* sign bit */
@@ -774,8 +775,8 @@ _mesa_half_to_float(GLhalfARB val)
       flt_m = m << 13;
    }
 
-   flt = (flt_s << 31) | (flt_e << 23) | flt_m;
-   result = *((float *) (void *) &flt);
+   fi.i = (flt_s << 31) | (flt_e << 23) | flt_m;
+   result = fi.f;
    return result;
 }
 
diff --git a/src/mesa/main/macros.h b/src/mesa/main/macros.h
index 3d9a1aba98d..f0ea463fb92 100644
--- a/src/mesa/main/macros.h
+++ b/src/mesa/main/macros.h
@@ -202,17 +202,12 @@ do {                                \
 #endif
 
 /**
- * Copy a 4-element float vector (avoid using FPU registers)
- * XXX Could use two 64-bit moves on 64-bit systems
+ * Copy a 4-element float vector
+ * memcpy seems to be most efficient
  */
 #define COPY_4FV( DST, SRC )                  \
 do {                                          \
-   const GLuint *_s = (const GLuint *) (SRC); \
-   GLuint *_d = (GLuint *) (DST);             \
-   _d[0] = _s[0];                             \
-   _d[1] = _s[1];                             \
-   _d[2] = _s[2];                             \
-   _d[3] = _s[3];                             \
+   _mesa_memcpy(DST, SRC, sizeof(GLfloat) * 4);       \
 } while (0)
 
 /** Copy \p SZ elements into a 4-element vector */
diff --git a/src/mesa/shader/prog_execute.c b/src/mesa/shader/prog_execute.c
index 192d39aed1e..038ce0b4f00 100644
--- a/src/mesa/shader/prog_execute.c
+++ b/src/mesa/shader/prog_execute.c
@@ -54,8 +54,18 @@
  * Set x to positive or negative infinity.
  */
 #if defined(USE_IEEE) || defined(_WIN32)
-#define SET_POS_INFINITY(x)  ( *((GLuint *) (void *)&x) = 0x7F800000 )
-#define SET_NEG_INFINITY(x)  ( *((GLuint *) (void *)&x) = 0xFF800000 )
+#define SET_POS_INFINITY(x)                  \
+   do {                                      \
+         fi_type fi;                         \
+         fi.i = 0x7F800000;                  \
+         x = fi.f;                           \
+   } while (0)
+#define SET_NEG_INFINITY(x)                  \
+   do {                                      \
+         fi_type fi;                         \
+         fi.i = 0xFF800000;                  \
+         x = fi.f;                           \
+   } while (0)
 #elif defined(VMS)
 #define SET_POS_INFINITY(x)  x = __MAXFLOAT
 #define SET_NEG_INFINITY(x)  x = -__MAXFLOAT
@@ -1635,11 +1645,12 @@ _mesa_execute_program(GLcontext * ctx,
       case OPCODE_UP2H:        /* unpack two 16-bit floats */
          {
             GLfloat a[4], result[4];
-            const GLuint *rawBits = (const GLuint *) a;
+            fi_type fi;
             GLhalfNV hx, hy;
             fetch_vector1(&inst->SrcReg[0], machine, a);
-            hx = rawBits[0] & 0xffff;
-            hy = rawBits[0] >> 16;
+            fi.f = a[0];
+            hx = fi.i & 0xffff;
+            hy = fi.i >> 16;
             result[0] = result[2] = _mesa_half_to_float(hx);
             result[1] = result[3] = _mesa_half_to_float(hy);
             store_vector4(inst, machine, result);
@@ -1648,11 +1659,12 @@ _mesa_execute_program(GLcontext * ctx,
       case OPCODE_UP2US:       /* unpack two GLushorts */
          {
             GLfloat a[4], result[4];
-            const GLuint *rawBits = (const GLuint *) a;
+            fi_type fi;
             GLushort usx, usy;
             fetch_vector1(&inst->SrcReg[0], machine, a);
-            usx = rawBits[0] & 0xffff;
-            usy = rawBits[0] >> 16;
+            fi.f = a[0];
+            usx = fi.i & 0xffff;
+            usy = fi.i >> 16;
             result[0] = result[2] = usx * (1.0f / 65535.0f);
             result[1] = result[3] = usy * (1.0f / 65535.0f);
             store_vector4(inst, machine, result);
@@ -1661,24 +1673,26 @@ _mesa_execute_program(GLcontext * ctx,
       case OPCODE_UP4B:        /* unpack four GLbytes */
          {
             GLfloat a[4], result[4];
-            const GLuint *rawBits = (const GLuint *) a;
+            fi_type fi;
             fetch_vector1(&inst->SrcReg[0], machine, a);
-            result[0] = (((rawBits[0] >> 0) & 0xff) - 128) / 127.0F;
-            result[1] = (((rawBits[0] >> 8) & 0xff) - 128) / 127.0F;
-            result[2] = (((rawBits[0] >> 16) & 0xff) - 128) / 127.0F;
-            result[3] = (((rawBits[0] >> 24) & 0xff) - 128) / 127.0F;
+            fi.f = a[0];
+            result[0] = (((fi.i >> 0) & 0xff) - 128) / 127.0F;
+            result[1] = (((fi.i >> 8) & 0xff) - 128) / 127.0F;
+            result[2] = (((fi.i >> 16) & 0xff) - 128) / 127.0F;
+            result[3] = (((fi.i >> 24) & 0xff) - 128) / 127.0F;
             store_vector4(inst, machine, result);
          }
          break;
       case OPCODE_UP4UB:       /* unpack four GLubytes */
          {
             GLfloat a[4], result[4];
-            const GLuint *rawBits = (const GLuint *) a;
+            fi_type fi;
             fetch_vector1(&inst->SrcReg[0], machine, a);
-            result[0] = ((rawBits[0] >> 0) & 0xff) / 255.0F;
-            result[1] = ((rawBits[0] >> 8) & 0xff) / 255.0F;
-            result[2] = ((rawBits[0] >> 16) & 0xff) / 255.0F;
-            result[3] = ((rawBits[0] >> 24) & 0xff) / 255.0F;
+            fi.f = a[0];
+            result[0] = ((fi.i >> 0) & 0xff) / 255.0F;
+            result[1] = ((fi.i >> 8) & 0xff) / 255.0F;
+            result[2] = ((fi.i >> 16) & 0xff) / 255.0F;
+            result[3] = ((fi.i >> 24) & 0xff) / 255.0F;
             store_vector4(inst, machine, result);
          }
          break;
diff --git a/src/mesa/state_tracker/st_atom_pixeltransfer.c b/src/mesa/state_tracker/st_atom_pixeltransfer.c
index 4b35f59cc2f..6a5854e9ba5 100644
--- a/src/mesa/state_tracker/st_atom_pixeltransfer.c
+++ b/src/mesa/state_tracker/st_atom_pixeltransfer.c
@@ -162,12 +162,14 @@ load_color_map_texture(GLcontext *ctx, struct pipe_texture *pt)
     */
    for (i = 0; i < texSize; i++) {
       for (j = 0; j < texSize; j++) {
+         union util_color uc;
          int k = (i * texSize + j);
          ubyte r = ctx->PixelMaps.RtoR.Map8[j * rSize / texSize];
          ubyte g = ctx->PixelMaps.GtoG.Map8[i * gSize / texSize];
          ubyte b = ctx->PixelMaps.BtoB.Map8[j * bSize / texSize];
          ubyte a = ctx->PixelMaps.AtoA.Map8[i * aSize / texSize];
-         util_pack_color_ub(r, g, b, a, pt->format, dest + k);
+         util_pack_color_ub(r, g, b, a, pt->format, &uc);
+         *(dest + k) = uc.ui;
       }
    }