Merge remote branch 'origin/lp-binning'

Conflicts:
	src/gallium/auxiliary/util/u_dl.c
	src/gallium/auxiliary/util/u_time.h
	src/gallium/drivers/llvmpipe/lp_state_derived.c
	src/gallium/drivers/llvmpipe/lp_state_surface.c
	src/gallium/drivers/llvmpipe/lp_tex_cache.c
	src/gallium/drivers/llvmpipe/lp_tile_cache.c

7 files changed, 338 insertions, 15 deletions

diff --git a/src/gallium/auxiliary/os/os_thread.h b/src/gallium/auxiliary/os/os_thread.h
index 2da5fd77386..24a2309976a 100644
--- a/src/gallium/auxiliary/os/os_thread.h
+++ b/src/gallium/auxiliary/os/os_thread.h
@@ -27,7 +27,8 @@
 /**
  * @file
  * 
- * Thread, mutex, condition var and thread-specific data functions.
+ * Thread, mutex, condition variable, barrier, semaphore and
+ * thread-specific data functions.
  */
 
 
@@ -46,6 +47,8 @@
 
 #define PIPE_THREAD_HAVE_CONDVAR
 
+/* pipe_thread
+ */
 typedef pthread_t pipe_thread;
 
 #define PIPE_THREAD_ROUTINE( name, param ) \
@@ -69,8 +72,10 @@ static INLINE int pipe_thread_destroy( pipe_thread thread )
    return pthread_detach( thread );
 }
 
+
+/* pipe_mutex
+ */
 typedef pthread_mutex_t pipe_mutex;
-typedef pthread_cond_t pipe_condvar;
 
 #define pipe_static_mutex(mutex) \
    static pipe_mutex mutex = PTHREAD_MUTEX_INITIALIZER
@@ -87,6 +92,11 @@ typedef pthread_cond_t pipe_condvar;
 #define pipe_mutex_unlock(mutex) \
    (void) pthread_mutex_unlock(&(mutex))
 
+
+/* pipe_condvar
+ */
+typedef pthread_cond_t pipe_condvar;
+
 #define pipe_static_condvar(mutex) \
    static pipe_condvar mutex = PTHREAD_COND_INITIALIZER
 
@@ -106,10 +116,32 @@ typedef pthread_cond_t pipe_condvar;
   pthread_cond_broadcast(&(cond))
 
 
+/* pipe_barrier
+ */
+typedef pthread_barrier_t pipe_barrier;
+
+static INLINE void pipe_barrier_init(pipe_barrier *barrier, unsigned count)
+{
+   pthread_barrier_init(barrier, NULL, count);
+}
+
+static INLINE void pipe_barrier_destroy(pipe_barrier *barrier)
+{
+   pthread_barrier_destroy(barrier);
+}
+
+static INLINE void pipe_barrier_wait(pipe_barrier *barrier)
+{
+   pthread_barrier_wait(barrier);
+}
+
+
 #elif defined(PIPE_SUBSYSTEM_WINDOWS_USER)
 
 #include <windows.h>
 
+/* pipe_thread
+ */
 typedef HANDLE pipe_thread;
 
 #define PIPE_THREAD_ROUTINE( name, param ) \
@@ -135,6 +167,9 @@ static INLINE int pipe_thread_destroy( pipe_thread thread )
    return -1;
 }
 
+
+/* pipe_mutex
+ */
 typedef CRITICAL_SECTION pipe_mutex;
 
 #define pipe_static_mutex(mutex) \
@@ -152,15 +187,48 @@ typedef CRITICAL_SECTION pipe_mutex;
 #define pipe_mutex_unlock(mutex) \
    LeaveCriticalSection(&mutex)
 
-/* XXX: dummy definitions, make it compile */
 
+/* pipe_condvar (XXX FIX THIS)
+ */
 typedef unsigned pipe_condvar;
 
-#define pipe_condvar_init(condvar) \
-   (void) condvar
+#define pipe_condvar_init(cond) \
+   (void) cond
+
+#define pipe_condvar_destroy(cond) \
+   (void) cond
+
+#define pipe_condvar_wait(cond, mutex) \
+   (void) cond; (void) mutex
+
+#define pipe_condvar_signal(cond) \
+   (void) cond
+
+#define pipe_condvar_broadcast(cond) \
+   (void) cond
+
+
+/* pipe_barrier (XXX FIX THIS)
+ */
+typedef unsigned pipe_barrier;
+
+static INLINE void pipe_barrier_init(pipe_barrier *barrier, unsigned count)
+{
+   /* XXX we could implement barriers with a mutex and condition var */
+   assert(0);
+}
+
+static INLINE void pipe_barrier_destroy(pipe_barrier *barrier)
+{
+   assert(0);
+}
+
+static INLINE void pipe_barrier_wait(pipe_barrier *barrier)
+{
+   assert(0);
+}
+
 
-#define pipe_condvar_broadcast(condvar) \
-   (void) condvar
 
 #else
 
@@ -188,6 +256,7 @@ static INLINE int pipe_thread_destroy( pipe_thread thread )
 
 typedef unsigned pipe_mutex;
 typedef unsigned pipe_condvar;
+typedef unsigned pipe_barrier;
 
 #define pipe_static_mutex(mutex) \
    static pipe_mutex mutex = 0
@@ -223,9 +292,77 @@ typedef unsigned pipe_condvar;
    (void) condvar
 
 
+static INLINE void pipe_barrier_init(pipe_barrier *barrier, unsigned count)
+{
+   /* XXX we could implement barriers with a mutex and condition var */
+   assert(0);
+}
+
+static INLINE void pipe_barrier_destroy(pipe_barrier *barrier)
+{
+   assert(0);
+}
+
+static INLINE void pipe_barrier_wait(pipe_barrier *barrier)
+{
+   assert(0);
+}
+
+
+
 #endif  /* PIPE_OS_? */
 
 
+/*
+ * Semaphores
+ */
+
+typedef struct
+{
+   pipe_mutex mutex;
+   pipe_condvar cond;
+   int counter;
+} pipe_semaphore;
+
+
+static INLINE void
+pipe_semaphore_init(pipe_semaphore *sema, int init_val)
+{
+   pipe_mutex_init(sema->mutex);
+   pipe_condvar_init(sema->cond);
+   sema->counter = init_val;
+}
+
+static INLINE void
+pipe_semaphore_destroy(pipe_semaphore *sema)
+{
+   pipe_mutex_destroy(sema->mutex);
+   pipe_condvar_destroy(sema->cond);
+}
+
+/** Signal/increment semaphore counter */
+static INLINE void
+pipe_semaphore_signal(pipe_semaphore *sema)
+{
+   pipe_mutex_lock(sema->mutex);
+   sema->counter++;
+   pipe_condvar_signal(sema->cond);
+   pipe_mutex_unlock(sema->mutex);
+}
+
+/** Wait for semaphore counter to be greater than zero */
+static INLINE void
+pipe_semaphore_wait(pipe_semaphore *sema)
+{
+   pipe_mutex_lock(sema->mutex);
+   while (sema->counter <= 0) {
+      pipe_condvar_wait(sema->cond, sema->mutex);
+   }
+   sema->counter--;
+   pipe_mutex_unlock(sema->mutex);
+}
+
+
 
 /*
  * Thread-specific data.
diff --git a/src/gallium/auxiliary/util/u_debug.c b/src/gallium/auxiliary/util/u_debug.c
index a8d18333d89..4821b8a1434 100644
--- a/src/gallium/auxiliary/util/u_debug.c
+++ b/src/gallium/auxiliary/util/u_debug.c
@@ -440,6 +440,14 @@ const char *u_prim_name( unsigned prim )
 
 
 #ifdef DEBUG
+/**
+ * Dump an image to a .raw or .ppm file (depends on OS).
+ * \param format  PIPE_FORMAT_x
+ * \param cpp  bytes per pixel
+ * \param width  width in pixels
+ * \param height height in pixels
+ * \param stride  row stride in bytes
+ */
 void debug_dump_image(const char *prefix,
                       unsigned format, unsigned cpp,
                       unsigned width, unsigned height,
@@ -481,6 +489,52 @@ void debug_dump_image(const char *prefix,
    }
       
    EngUnmapFile(iFile);
+#elif defined(PIPE_OS_UNIX)
+   /* write a ppm file */
+   char filename[256];
+   FILE *f;
+
+   util_snprintf(filename, sizeof(filename), "%s.ppm", prefix);
+
+   f = fopen(filename, "w");
+   if (f) {
+      int i, x, y;
+      int r, g, b;
+      const uint8_t *ptr = (uint8_t *) data;
+
+      /* XXX this is a hack */
+      switch (format) {
+      case PIPE_FORMAT_A8R8G8B8_UNORM:
+         r = 2;
+         g = 1;
+         b = 0;
+         break;
+      default:
+         r = 0;
+         g = 1;
+         b = 1;
+      }
+
+      fprintf(f, "P6\n");
+      fprintf(f, "# ppm-file created by osdemo.c\n");
+      fprintf(f, "%i %i\n", width, height);
+      fprintf(f, "255\n");
+      fclose(f);
+
+      f = fopen(filename, "ab");  /* reopen in binary append mode */
+      for (y = 0; y < height; y++) {
+         for (x = 0; x < width; x++) {
+            i = y * stride + x * cpp;
+            fputc(ptr[i + r], f); /* write red */
+            fputc(ptr[i + g], f); /* write green */
+            fputc(ptr[i + b], f); /* write blue */
+         }
+      }
+      fclose(f);
+   }
+   else {
+      fprintf(stderr, "Can't open %s for writing\n", filename);
+   }
 #endif
 }
 
@@ -521,6 +575,27 @@ error:
 }
 
 
+void debug_dump_texture(const char *prefix,
+                        struct pipe_texture *texture)
+{
+   struct pipe_surface *surface;
+   struct pipe_screen *screen;
+
+   if (!texture)
+      return;
+
+   screen = texture->screen;
+
+   /* XXX for now, just dump image for face=0, level=0 */
+   surface = screen->get_tex_surface(screen, texture, 0, 0, 0,
+                                     PIPE_TEXTURE_USAGE_SAMPLER);
+   if (surface) {
+      debug_dump_surface(prefix, surface);
+      screen->tex_surface_destroy(surface);
+   }
+}
+
+
 #pragma pack(push,2)
 struct bmp_file_header {
    uint16_t bfType;
diff --git a/src/gallium/auxiliary/util/u_debug.h b/src/gallium/auxiliary/util/u_debug.h
index eadc08fe2a0..efcf065d276 100644
--- a/src/gallium/auxiliary/util/u_debug.h
+++ b/src/gallium/auxiliary/util/u_debug.h
@@ -314,6 +314,8 @@ debug_memory_end(unsigned long beginning);
 #ifdef DEBUG
 struct pipe_surface;
 struct pipe_transfer;
+struct pipe_texture;
+
 void debug_dump_image(const char *prefix,
                       unsigned format, unsigned cpp,
                       unsigned width, unsigned height,
@@ -321,6 +323,8 @@ void debug_dump_image(const char *prefix,
                       const void *data);
 void debug_dump_surface(const char *prefix,
                         struct pipe_surface *surface);   
+void debug_dump_texture(const char *prefix,
+                        struct pipe_texture *texture);
 void debug_dump_surface_bmp(const char *filename,
                             struct pipe_surface *surface);
 void debug_dump_transfer_bmp(const char *filename,
diff --git a/src/gallium/auxiliary/util/u_ringbuffer.c b/src/gallium/auxiliary/util/u_ringbuffer.c
index 95d45ebb71f..648b105b137 100644
--- a/src/gallium/auxiliary/util/u_ringbuffer.c
+++ b/src/gallium/auxiliary/util/u_ringbuffer.c
@@ -53,11 +53,22 @@ void util_ringbuffer_destroy( struct util_ringbuffer *ring )
    FREE(ring);
 }
 
+/**
+ * Return number of free entries in the ring
+ */
 static INLINE unsigned util_ringbuffer_space( const struct util_ringbuffer *ring )
 {
    return (ring->tail - (ring->head + 1)) & ring->mask;
 }
 
+/**
+ * Is the ring buffer empty?
+ */
+static INLINE boolean util_ringbuffer_empty( const struct util_ringbuffer *ring )
+{
+   return util_ringbuffer_space(ring) == ring->mask;
+}
+
 void util_ringbuffer_enqueue( struct util_ringbuffer *ring,
                               const struct util_packet *packet )
 {
@@ -67,6 +78,10 @@ void util_ringbuffer_enqueue( struct util_ringbuffer *ring,
     */
    pipe_mutex_lock(ring->mutex);
 
+   /* make sure we don't request an impossible amount of space
+    */
+   assert(packet->dwords <= ring->mask);
+
    /* Wait for free space:
     */
    while (util_ringbuffer_space(ring) < packet->dwords)
@@ -104,14 +119,14 @@ enum pipe_error util_ringbuffer_dequeue( struct util_ringbuffer *ring,
     */
    pipe_mutex_lock(ring->mutex);
 
-   /* Wait for free space:
+   /* Get next ring entry:
     */
    if (wait) {
-      while (util_ringbuffer_space(ring) == 0)
+      while (util_ringbuffer_empty(ring))
          pipe_condvar_wait(ring->change, ring->mutex);
    }
    else {
-      if (util_ringbuffer_space(ring) == 0) {
+      if (util_ringbuffer_empty(ring)) {
          ret = PIPE_ERROR_OUT_OF_MEMORY;
          goto out;
       }
diff --git a/src/gallium/auxiliary/util/u_surface.c b/src/gallium/auxiliary/util/u_surface.c
index 6053c111e34..c9f1c9c210f 100644
--- a/src/gallium/auxiliary/util/u_surface.c
+++ b/src/gallium/auxiliary/util/u_surface.c
@@ -37,6 +37,7 @@
 #include "pipe/p_defines.h"
 #include "util/u_inlines.h"
 
+#include "util/u_memory.h"
 #include "util/u_surface.h"
 
 
@@ -111,3 +112,73 @@ util_destroy_rgba_surface(struct pipe_texture *texture,
    pipe_texture_reference(&texture, NULL);
 }
 
+
+
+/**
+ * Compare pipe_framebuffer_state objects.
+ * \return TRUE if same, FALSE if different
+ */
+boolean
+util_framebuffer_state_equal(const struct pipe_framebuffer_state *dst,
+                             const struct pipe_framebuffer_state *src)
+{
+   unsigned i;
+
+   if (dst->width != src->width ||
+       dst->height != src->height)
+      return FALSE;
+
+   for (i = 0; i < Elements(src->cbufs); i++) {
+      if (dst->cbufs[i] != src->cbufs[i]) {
+         return FALSE;
+      }
+   }
+
+   if (dst->nr_cbufs != src->nr_cbufs) {
+      return FALSE;
+   }
+
+   if (dst->zsbuf != src->zsbuf) {
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+
+/**
+ * Copy framebuffer state from src to dst, updating refcounts.
+ */
+void
+util_copy_framebuffer_state(struct pipe_framebuffer_state *dst,
+                            const struct pipe_framebuffer_state *src)
+{
+   unsigned i;
+
+   dst->width = src->width;
+   dst->height = src->height;
+
+   for (i = 0; i < Elements(src->cbufs); i++) {
+      pipe_surface_reference(&dst->cbufs[i], src->cbufs[i]);
+   }
+
+   dst->nr_cbufs = src->nr_cbufs;
+
+   pipe_surface_reference(&dst->zsbuf, src->zsbuf);
+}
+
+
+void
+util_unreference_framebuffer_state(struct pipe_framebuffer_state *fb)
+{
+   unsigned i;
+
+   for (i = 0; i < fb->nr_cbufs; i++) {
+      pipe_surface_reference(&fb->cbufs[i], NULL);
+   }
+
+   pipe_surface_reference(&fb->zsbuf, NULL);
+
+   fb->width = fb->height = 0;
+   fb->nr_cbufs = 0;
+}
diff --git a/src/gallium/auxiliary/util/u_surface.h b/src/gallium/auxiliary/util/u_surface.h
index ce84ed7ad06..3c60df2c3e5 100644
--- a/src/gallium/auxiliary/util/u_surface.h
+++ b/src/gallium/auxiliary/util/u_surface.h
@@ -30,11 +30,7 @@
 
 
 #include "pipe/p_compiler.h"
-
-
-struct pipe_screen;
-struct pipe_texture;
-struct pipe_surface;
+#include "pipe/p_state.h"
 
 
 /**
@@ -66,4 +62,17 @@ util_destroy_rgba_surface(struct pipe_texture *texture,
                           struct pipe_surface *surface);
 
 
+extern boolean
+util_framebuffer_state_equal(const struct pipe_framebuffer_state *dst,
+                             const struct pipe_framebuffer_state *src);
+
+extern void
+util_copy_framebuffer_state(struct pipe_framebuffer_state *dst,
+                            const struct pipe_framebuffer_state *src);
+
+
+extern void
+util_unreference_framebuffer_state(struct pipe_framebuffer_state *fb);
+
+
 #endif /* U_SURFACE_H */
diff --git a/src/gallium/auxiliary/util/u_time.h b/src/gallium/auxiliary/util/u_time.h
index 7580ac0de4c..15899c2c884 100644
--- a/src/gallium/auxiliary/util/u_time.h
+++ b/src/gallium/auxiliary/util/u_time.h
@@ -67,6 +67,9 @@ util_time_get(struct util_time *t)
 }
 
 
+/**
+ * Return t2 = t1 + usecs
+ */
 PIPE_DEPRECATED
 static INLINE void
 util_time_add(const struct util_time *t1,
@@ -77,6 +80,9 @@ util_time_add(const struct util_time *t1,
 }
 
 
+/**
+ * Return difference between times, in microseconds
+ */
 PIPE_DEPRECATED
 static INLINE int64_t
 util_time_diff(const struct util_time *t1, 
@@ -105,6 +111,9 @@ _util_time_compare(const struct util_time *t1,
 }
 
 
+/**
+ * Returns non-zero when the timeout expires.
+ */
 PIPE_DEPRECATED
 static INLINE boolean
 util_time_timeout(const struct util_time *start, 
@@ -115,6 +124,9 @@ util_time_timeout(const struct util_time *start,
 }
 
 
+/**
+ * Return current time in microseconds
+ */
 PIPE_DEPRECATED
 static INLINE int64_t
 util_time_micros(void)