X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fdrivers%2Fdri%2Fi915%2Fintel_regions.c;h=c9b776d1a0f094e25db95bf73bee512a997a875f;hb=0febd0ecfd1e2a36381ab7793811b9c7891ed82f;hp=7799fdbb116b2218e621bd78155832a946b2592a;hpb=dc1264970e585c0939606922b3e964db4b9c28b3;p=mesa.git

diff --git a/src/mesa/drivers/dri/i915/intel_regions.c b/src/mesa/drivers/dri/i915/intel_regions.c
index 7799fdbb116..c9b776d1a0f 100644
--- a/src/mesa/drivers/dri/i915/intel_regions.c
+++ b/src/mesa/drivers/dri/i915/intel_regions.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2006 VMware, Inc.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -18,7 +18,7 @@
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
@@ -39,445 +39,315 @@
  * last moment.
  */
 
+#include <sys/ioctl.h>
+#include <errno.h>
+
+#include "main/hash.h"
 #include "intel_context.h"
 #include "intel_regions.h"
 #include "intel_blit.h"
 #include "intel_buffer_objects.h"
-#include "dri_bufmgr.h"
-#include "intel_bufmgr_ttm.h"
+#include "intel_bufmgr.h"
 #include "intel_batchbuffer.h"
 
 #define FILE_DEBUG_FLAG DEBUG_REGION
 
-void
-intel_region_idle(intelScreenPrivate *intelScreen, struct intel_region *region)
+/* This should be set to the maximum backtrace size desired.
+ * Set it to 0 to disable backtrace debugging.
+ */
+#define DEBUG_BACKTRACE_SIZE 0
+
+#if DEBUG_BACKTRACE_SIZE == 0
+/* Use the standard debug output */
+#define _DBG(...) DBG(__VA_ARGS__)
+#else
+/* Use backtracing debug output */
+#define _DBG(...) {debug_backtrace(); DBG(__VA_ARGS__);}
+
+/* Backtracing debug support */
+#include <execinfo.h>
+
+static void
+debug_backtrace(void)
 {
-   DBG("%s\n", __FUNCTION__);
-   /* XXX: Using this function is likely bogus -- it ought to only have been
-    * used before a map, anyway, but leave this cheap implementation of it
-    * for now.
-    */
-   if (region && region->buffer) {
-      /* Mapping it for read will ensure that any acceleration to the region
-       * would have landed already.
-       */
-      dri_bo_map(region->buffer, GL_TRUE);
-      dri_bo_unmap(region->buffer);
+   void *trace[DEBUG_BACKTRACE_SIZE];
+   char **strings = NULL;
+   int traceSize;
+   register int i;
+
+   traceSize = backtrace(trace, DEBUG_BACKTRACE_SIZE);
+   strings = backtrace_symbols(trace, traceSize);
+   if (strings == NULL) {
+      DBG("no backtrace:");
+      return;
    }
-}
 
-/* XXX: Thread safety?
- */
-GLubyte *
-intel_region_map(intelScreenPrivate *intelScreen, struct intel_region *region)
-{
-   DBG("%s\n", __FUNCTION__);
-   if (!region->map_refcount++) {
-      if (region->pbo)
-         intel_region_cow(intelScreen, region);
+   /* Spit out all the strings with a colon separator.  Ignore
+    * the first, since we don't really care about the call
+    * to debug_backtrace() itself.  Skip until the final "/" in
+    * the trace to avoid really long lines.
+    */
+   for (i = 1; i < traceSize; i++) {
+      char *p = strings[i], *slash = strings[i];
+      while (*p) {
+         if (*p++ == '/') {
+            slash = p;
+         }
+      }
 
-      dri_bo_map(region->buffer, GL_TRUE);
-      region->map = region->buffer->virtual;
+      DBG("%s:", slash);
    }
 
-   return region->map;
+   /* Free up the memory, and we're done */
+   free(strings);
 }
 
-void
-intel_region_unmap(intelScreenPrivate *intelScreen, struct intel_region *region)
-{
-   DBG("%s\n", __FUNCTION__);
-   if (!--region->map_refcount) {
-      dri_bo_unmap(region->buffer);
-      region->map = NULL;
-   }
-}
+#endif
 
-struct intel_region *
-intel_region_alloc(intelScreenPrivate *intelScreen,
-                   GLuint cpp, GLuint pitch, GLuint height)
+static struct intel_region *
+intel_region_alloc_internal(struct intel_screen *screen,
+			    GLuint cpp,
+			    GLuint width, GLuint height, GLuint pitch,
+			    uint32_t tiling, drm_intel_bo *buffer)
 {
-   struct intel_region *region = calloc(sizeof(*region), 1);
+   struct intel_region *region;
 
-   DBG("%s\n", __FUNCTION__);
+   region = calloc(sizeof(*region), 1);
+   if (region == NULL)
+      return region;
 
    region->cpp = cpp;
+   region->width = width;
+   region->height = height;
    region->pitch = pitch;
-   region->height = height;     /* needed? */
    region->refcount = 1;
+   region->bo = buffer;
+   region->tiling = tiling;
 
-   region->buffer = dri_bo_alloc(intelScreen->bufmgr, "region",
-				 pitch * cpp * height, 64, DRM_BO_FLAG_MEM_TT);
+   _DBG("%s <-- %p\n", __func__, region);
    return region;
 }
 
-void
-intel_region_reference(struct intel_region **dst, struct intel_region *src)
+struct intel_region *
+intel_region_alloc(struct intel_screen *screen,
+		   uint32_t tiling,
+                   GLuint cpp, GLuint width, GLuint height,
+		   bool expect_accelerated_upload)
 {
-   assert(*dst == NULL);
-   if (src) {
-      src->refcount++;
-      *dst = src;
+   drm_intel_bo *buffer;
+   unsigned long flags = 0;
+   unsigned long aligned_pitch;
+   struct intel_region *region;
+
+   if (expect_accelerated_upload)
+      flags |= BO_ALLOC_FOR_RENDER;
+
+   buffer = drm_intel_bo_alloc_tiled(screen->bufmgr, "region",
+				     width, height, cpp,
+				     &tiling, &aligned_pitch, flags);
+   if (buffer == NULL)
+      return NULL;
+
+   region = intel_region_alloc_internal(screen, cpp, width, height,
+                                        aligned_pitch, tiling, buffer);
+   if (region == NULL) {
+      drm_intel_bo_unreference(buffer);
+      return NULL;
    }
+
+   return region;
 }
 
-void
-intel_region_release(struct intel_region **region)
+bool
+intel_region_flink(struct intel_region *region, uint32_t *name)
 {
-   if (!*region)
-      return;
-
-   DBG("%s %d\n", __FUNCTION__, (*region)->refcount - 1);
-
-   ASSERT((*region)->refcount > 0);
-   (*region)->refcount--;
+   if (region->name == 0) {
+      if (drm_intel_bo_flink(region->bo, &region->name))
+	 return false;
+   }
 
-   if ((*region)->refcount == 0) {
-      assert((*region)->map_refcount == 0);
+   *name = region->name;
 
-      if ((*region)->pbo)
-	 (*region)->pbo->region = NULL;
-      (*region)->pbo = NULL;
-      dri_bo_unreference((*region)->buffer);
-      free(*region);
-   }
-   *region = NULL;
+   return true;
 }
 
-
 struct intel_region *
-intel_region_create_static(intelScreenPrivate *intelScreen,
-                           GLuint mem_type,
-			   unsigned int bo_handle,
-                           GLuint offset,
-                           void *virtual,
-                           GLuint cpp, GLuint pitch, GLuint height)
+intel_region_alloc_for_handle(struct intel_screen *screen,
+			      GLuint cpp,
+			      GLuint width, GLuint height, GLuint pitch,
+			      GLuint handle, const char *name)
 {
-   struct intel_region *region = calloc(sizeof(*region), 1);
-   DBG("%s\n", __FUNCTION__);
-
-   region->cpp = cpp;
-   region->pitch = pitch;
-   region->height = height;     /* needed? */
-   region->refcount = 1;
+   struct intel_region *region;
+   drm_intel_bo *buffer;
+   int ret;
+   uint32_t bit_6_swizzle, tiling;
+
+   buffer = intel_bo_gem_create_from_name(screen->bufmgr, name, handle);
+   if (buffer == NULL)
+      return NULL;
+   ret = drm_intel_bo_get_tiling(buffer, &tiling, &bit_6_swizzle);
+   if (ret != 0) {
+      fprintf(stderr, "Couldn't get tiling of buffer %d (%s): %s\n",
+	      handle, name, strerror(-ret));
+      drm_intel_bo_unreference(buffer);
+      return NULL;
+   }
 
-   if (intelScreen->ttm) {
-      assert(bo_handle != -1);
-      region->buffer = intel_ttm_bo_create_from_handle(intelScreen->bufmgr,
-						     "static region",
-						     bo_handle);
-   } else {
-      region->buffer = dri_bo_alloc_static(intelScreen->bufmgr,
-					   "static region",
-					   offset, pitch * cpp * height,
-					   virtual,
-					   DRM_BO_FLAG_MEM_TT);
+   region = intel_region_alloc_internal(screen, cpp,
+					width, height, pitch, tiling, buffer);
+   if (region == NULL) {
+      drm_intel_bo_unreference(buffer);
+      return NULL;
    }
 
+   region->name = handle;
+
    return region;
 }
 
-
-
-void
-intel_region_update_static(intelScreenPrivate *intelScreen,
-			   struct intel_region *region,
-                           GLuint mem_type,
-			   unsigned int bo_handle,
-                           GLuint offset,
-                           void *virtual,
-                           GLuint cpp, GLuint pitch, GLuint height)
+struct intel_region *
+intel_region_alloc_for_fd(struct intel_screen *screen,
+                          GLuint cpp,
+                          GLuint width, GLuint height, GLuint pitch,
+                          GLuint size,
+                          int fd, const char *name)
 {
-   DBG("%s\n", __FUNCTION__);
-
-   region->cpp = cpp;
-   region->pitch = pitch;
-   region->height = height;     /* needed? */
-
-   /*
-    * We use a "shared" buffer type to indicate buffers created and
-    * shared by others.
-    */
-
-   dri_bo_unreference(region->buffer);
-   if (intelScreen->ttm) {
-      assert(bo_handle != -1);
-      region->buffer = intel_ttm_bo_create_from_handle(intelScreen->bufmgr,
-						     "static region",
-						     bo_handle);
-   } else {
-      region->buffer = dri_bo_alloc_static(intelScreen->bufmgr,
-					   "static region",
-					   offset, pitch * cpp * height,
-					   virtual,
-					   DRM_BO_FLAG_MEM_TT);
+   struct intel_region *region;
+   drm_intel_bo *buffer;
+   int ret;
+   uint32_t bit_6_swizzle, tiling;
+
+   buffer = drm_intel_bo_gem_create_from_prime(screen->bufmgr, fd, size);
+   if (buffer == NULL)
+      return NULL;
+   ret = drm_intel_bo_get_tiling(buffer, &tiling, &bit_6_swizzle);
+   if (ret != 0) {
+      fprintf(stderr, "Couldn't get tiling of buffer (%s): %s\n",
+	      name, strerror(-ret));
+      drm_intel_bo_unreference(buffer);
+      return NULL;
    }
-}
-
 
-
-/*
- * XXX Move this into core Mesa?
- */
-static void
-_mesa_copy_rect(GLubyte * dst,
-                GLuint cpp,
-                GLuint dst_pitch,
-                GLuint dst_x,
-                GLuint dst_y,
-                GLuint width,
-                GLuint height,
-                const GLubyte * src,
-                GLuint src_pitch, GLuint src_x, GLuint src_y)
-{
-   GLuint i;
-
-   dst_pitch *= cpp;
-   src_pitch *= cpp;
-   dst += dst_x * cpp;
-   src += src_x * cpp;
-   dst += dst_y * dst_pitch;
-   src += src_y * dst_pitch;
-   width *= cpp;
-
-   if (width == dst_pitch && width == src_pitch)
-      memcpy(dst, src, height * width);
-   else {
-      for (i = 0; i < height; i++) {
-         memcpy(dst, src, width);
-         dst += dst_pitch;
-         src += src_pitch;
-      }
+   region = intel_region_alloc_internal(screen, cpp,
+					width, height, pitch, tiling, buffer);
+   if (region == NULL) {
+      drm_intel_bo_unreference(buffer);
+      return NULL;
    }
-}
 
+   return region;
+}
 
-/* Upload data to a rectangular sub-region.  Lots of choices how to do this:
- *
- * - memcpy by span to current destination
- * - upload data as new buffer and blit
- *
- * Currently always memcpy.
- */
 void
-intel_region_data(intelScreenPrivate *intelScreen,
-                  struct intel_region *dst,
-                  GLuint dst_offset,
-                  GLuint dstx, GLuint dsty,
-                  const void *src, GLuint src_pitch,
-                  GLuint srcx, GLuint srcy, GLuint width, GLuint height)
+intel_region_reference(struct intel_region **dst, struct intel_region *src)
 {
-   struct intel_context *intel = intelScreenContext(intelScreen);
+   _DBG("%s: %p(%d) -> %p(%d)\n", __func__,
+	*dst, *dst ? (*dst)->refcount : 0, src, src ? src->refcount : 0);
 
-   DBG("%s\n", __FUNCTION__);
-
-   if (intel == NULL)
-      return;
+   if (src != *dst) {
+      if (*dst)
+	 intel_region_release(dst);
 
-   if (dst->pbo) {
-      if (dstx == 0 &&
-          dsty == 0 && width == dst->pitch && height == dst->height)
-         intel_region_release_pbo(intelScreen, dst);
-      else
-         intel_region_cow(intelScreen, dst);
+      if (src)
+         src->refcount++;
+      *dst = src;
    }
-
-
-   LOCK_HARDWARE(intel);
-
-   _mesa_copy_rect(intel_region_map(intelScreen, dst) + dst_offset,
-                   dst->cpp,
-                   dst->pitch,
-                   dstx, dsty, width, height, src, src_pitch, srcx, srcy);
-
-   intel_region_unmap(intelScreen, dst);
-
-   UNLOCK_HARDWARE(intel);
-
 }
 
-/* Copy rectangular sub-regions. Need better logic about when to
- * push buffers into AGP - will currently do so whenever possible.
- */
 void
-intel_region_copy(intelScreenPrivate *intelScreen,
-                  struct intel_region *dst,
-                  GLuint dst_offset,
-                  GLuint dstx, GLuint dsty,
-                  struct intel_region *src,
-                  GLuint src_offset,
-                  GLuint srcx, GLuint srcy, GLuint width, GLuint height)
+intel_region_release(struct intel_region **region_handle)
 {
-   struct intel_context *intel = intelScreenContext(intelScreen);
-
-   DBG("%s\n", __FUNCTION__);
+   struct intel_region *region = *region_handle;
 
-   if (intel == NULL)
+   if (region == NULL) {
+      _DBG("%s NULL\n", __func__);
       return;
-
-   if (dst->pbo) {
-      if (dstx == 0 &&
-          dsty == 0 && width == dst->pitch && height == dst->height)
-         intel_region_release_pbo(intelScreen, dst);
-      else
-         intel_region_cow(intelScreen, dst);
    }
 
-   assert(src->cpp == dst->cpp);
-
-   intelEmitCopyBlit(intel,
-                     dst->cpp,
-                     src->pitch, src->buffer, src_offset,
-                     dst->pitch, dst->buffer, dst_offset,
-                     srcx, srcy, dstx, dsty, width, height,
-		     GL_COPY);
-}
-
-/* Fill a rectangular sub-region.  Need better logic about when to
- * push buffers into AGP - will currently do so whenever possible.
- */
-void
-intel_region_fill(intelScreenPrivate *intelScreen,
-                  struct intel_region *dst,
-                  GLuint dst_offset,
-                  GLuint dstx, GLuint dsty,
-                  GLuint width, GLuint height, GLuint color)
-{
-   struct intel_context *intel = intelScreenContext(intelScreen);
+   _DBG("%s %p %d\n", __func__, region, region->refcount - 1);
 
-   DBG("%s\n", __FUNCTION__);
+   assert(region->refcount > 0);
+   region->refcount--;
 
-   if (intel == NULL)
-      return;   
+   if (region->refcount == 0) {
+      drm_intel_bo_unreference(region->bo);
 
-   if (dst->pbo) {
-      if (dstx == 0 &&
-          dsty == 0 && width == dst->pitch && height == dst->height)
-         intel_region_release_pbo(intelScreen, dst);
-      else
-         intel_region_cow(intelScreen, dst);
+      free(region);
    }
-
-   intelEmitFillBlit(intel,
-                     dst->cpp,
-                     dst->pitch, dst->buffer, dst_offset,
-                     dstx, dsty, width, height, color);
+   *region_handle = NULL;
 }
 
-/* Attach to a pbo, discarding our data.  Effectively zero-copy upload
- * the pbo's data.
+/**
+ * This function computes masks that may be used to select the bits of the X
+ * and Y coordinates that indicate the offset within a tile.  If the region is
+ * untiled, the masks are set to 0.
  */
 void
-intel_region_attach_pbo(intelScreenPrivate *intelScreen,
-                        struct intel_region *region,
-                        struct intel_buffer_object *pbo)
+intel_region_get_tile_masks(struct intel_region *region,
+                            uint32_t *mask_x, uint32_t *mask_y,
+                            bool map_stencil_as_y_tiled)
 {
-   if (region->pbo == pbo)
-      return;
-
-   /* If there is already a pbo attached, break the cow tie now.
-    * Don't call intel_region_release_pbo() as that would
-    * unnecessarily allocate a new buffer we would have to immediately
-    * discard.
-    */
-   if (region->pbo) {
-      region->pbo->region = NULL;
-      region->pbo = NULL;
+   int cpp = region->cpp;
+   uint32_t tiling = region->tiling;
+
+   if (map_stencil_as_y_tiled)
+      tiling = I915_TILING_Y;
+
+   switch (tiling) {
+   default:
+      assert(false);
+   case I915_TILING_NONE:
+      *mask_x = *mask_y = 0;
+      break;
+   case I915_TILING_X:
+      *mask_x = 512 / cpp - 1;
+      *mask_y = 7;
+      break;
+   case I915_TILING_Y:
+      *mask_x = 128 / cpp - 1;
+      *mask_y = 31;
+      break;
    }
-
-   if (region->buffer) {
-      dri_bo_unreference(region->buffer);
-      region->buffer = NULL;
-   }
-
-   region->pbo = pbo;
-   region->pbo->region = region;
-   dri_bo_reference(pbo->buffer);
-   region->buffer = pbo->buffer;
 }
 
-
-/* Break the COW tie to the pbo and allocate a new buffer.
- * The pbo gets to keep the data.
+/**
+ * Compute the offset (in bytes) from the start of the region to the given x
+ * and y coordinate.  For tiled regions, caller must ensure that x and y are
+ * multiples of the tile size.
  */
-void
-intel_region_release_pbo(intelScreenPrivate *intelScreen,
-                         struct intel_region *region)
-{
-   assert(region->buffer == region->pbo->buffer);
-   region->pbo->region = NULL;
-   region->pbo = NULL;
-   dri_bo_unreference(region->buffer);
-   region->buffer = NULL;
-
-   region->buffer = dri_bo_alloc(intelScreen->bufmgr, "region",
-				 region->pitch * region->cpp * region->height,
-				 64, DRM_BO_FLAG_MEM_TT);
-}
-
-/* Break the COW tie to the pbo.  Both the pbo and the region end up
- * with a copy of the data.
- */
-void
-intel_region_cow(intelScreenPrivate *intelScreen, struct intel_region *region)
+uint32_t
+intel_region_get_aligned_offset(struct intel_region *region, uint32_t x,
+                                uint32_t y, bool map_stencil_as_y_tiled)
 {
-   struct intel_context *intel = intelScreenContext(intelScreen);
-   struct intel_buffer_object *pbo = region->pbo;
-
-   if (intel == NULL)
-      return;
-
-   intel_region_release_pbo(intelScreen, region);
-
-   assert(region->cpp * region->pitch * region->height == pbo->Base.Size);
-
-   DBG("%s (%d bytes)\n", __FUNCTION__, pbo->Base.Size);
-
-   /* Now blit from the texture buffer to the new buffer: 
-    */
-
-   intel_batchbuffer_flush(intel->batch);
-
-   if (!intel->locked) {
-      LOCK_HARDWARE(intel);
-      intelEmitCopyBlit(intel,
-			region->cpp,
-			region->pitch,
-			region->buffer, 0,
-			region->pitch,
-			pbo->buffer, 0,
-			0, 0, 0, 0, 
-			region->pitch, region->height,
-			GL_COPY);
-      
-      intel_batchbuffer_flush(intel->batch);
-      UNLOCK_HARDWARE(intel);
-   }
-   else {
-      intelEmitCopyBlit(intel,
-			region->cpp,
-			region->pitch,
-			region->buffer, 0,
-			region->pitch,
-			pbo->buffer, 0,
-			0, 0, 0, 0, 
-			region->pitch, region->height,
-			GL_COPY);
-      
-      intel_batchbuffer_flush(intel->batch);
+   int cpp = region->cpp;
+   uint32_t pitch = region->pitch;
+   uint32_t tiling = region->tiling;
+
+   if (map_stencil_as_y_tiled) {
+      tiling = I915_TILING_Y;
+
+      /* When mapping a W-tiled stencil buffer as Y-tiled, each 64-high W-tile
+       * gets transformed into a 32-high Y-tile.  Accordingly, the pitch of
+       * the resulting region is twice the pitch of the original region, since
+       * each row in the Y-tiled view corresponds to two rows in the actual
+       * W-tiled surface.  So we need to correct the pitch before computing
+       * the offsets.
+       */
+      pitch *= 2;
    }
-}
 
-dri_bo *
-intel_region_buffer(intelScreenPrivate *intelScreen,
-                    struct intel_region *region, GLuint flag)
-{
-   if (region->pbo) {
-      if (flag == INTEL_WRITE_PART)
-         intel_region_cow(intelScreen, region);
-      else if (flag == INTEL_WRITE_FULL)
-         intel_region_release_pbo(intelScreen, region);
+   switch (tiling) {
+   default:
+      assert(false);
+   case I915_TILING_NONE:
+      return y * pitch + x * cpp;
+   case I915_TILING_X:
+      assert((x % (512 / cpp)) == 0);
+      assert((y % 8) == 0);
+      return y * pitch + x / (512 / cpp) * 4096;
+   case I915_TILING_Y:
+      assert((x % (128 / cpp)) == 0);
+      assert((y % 32) == 0);
+      return y * pitch + x / (128 / cpp) * 4096;
    }
-
-   return region->buffer;
 }