intel: Add support for tiled textures.

This is about a 30% performance win in OA with high settings on my GM45,
and experiments with 915GM indicate that it'll be around a 20% win there.
Currently, 915-class hardware is seriously hurt by the fact that we use
fence regs to control the tiling even for 3D instructions that could live
without them, so we spend a bunch of time waiting on previous rendering in
order to pull fences off.  Thus, the texture_tiling driconf option defaults
off there for now.

11 files changed, 88 insertions, 20 deletions

diff --git a/src/mesa/drivers/dri/intel/intel_context.c b/src/mesa/drivers/dri/intel/intel_context.c
index ea43009f4c8..fa931d7f625 100644
--- a/src/mesa/drivers/dri/intel/intel_context.c
+++ b/src/mesa/drivers/dri/intel/intel_context.c
@@ -724,6 +724,8 @@ intelInitContext(struct intel_context *intel,
    else if (driQueryOptionb(&intel->optionCache, "force_s3tc_enable")) {
       _mesa_enable_extension(ctx, "GL_EXT_texture_compression_s3tc");
    }
+   intel->use_texture_tiling = driQueryOptionb(&intel->optionCache,
+					       "texture_tiling");
 
    intel->prim.primitive = ~0;
 
diff --git a/src/mesa/drivers/dri/intel/intel_context.h b/src/mesa/drivers/dri/intel/intel_context.h
index 810f3e62d98..4e45f1a91fb 100644
--- a/src/mesa/drivers/dri/intel/intel_context.h
+++ b/src/mesa/drivers/dri/intel/intel_context.h
@@ -305,6 +305,8 @@ struct intel_context
     */
    GLboolean is_front_buffer_rendering;
 
+   GLboolean use_texture_tiling;
+
    drm_clip_rect_t fboRect;     /**< cliprect for FBO rendering */
 
    int perf_boxes;
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c
index 04723a2f917..0ea413aee1d 100644
--- a/src/mesa/drivers/dri/intel/intel_fbo.c
+++ b/src/mesa/drivers/dri/intel/intel_fbo.c
@@ -217,7 +217,8 @@ intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
       DBG("Allocating %d x %d Intel RBO (pitch %d)\n", width,
 	  height, pitch);
 
-      irb->region = intel_region_alloc(intel, cpp, width, height, pitch,
+      irb->region = intel_region_alloc(intel, I915_TILING_NONE,
+				       cpp, width, height, pitch,
 				       GL_TRUE);
       if (!irb->region)
          return GL_FALSE;       /* out of memory? */
diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
index f3652720ece..0d34f28311e 100644
--- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
+++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
@@ -57,7 +57,8 @@ intel_miptree_create_internal(struct intel_context *intel,
 			      GLuint last_level,
 			      GLuint width0,
 			      GLuint height0,
-			      GLuint depth0, GLuint cpp, GLuint compress_byte)
+			      GLuint depth0, GLuint cpp, GLuint compress_byte,
+			      uint32_t tiling)
 {
    GLboolean ok;
    struct intel_mipmap_tree *mt = calloc(sizeof(*mt), 1);
@@ -81,11 +82,11 @@ intel_miptree_create_internal(struct intel_context *intel,
 
 #ifdef I915
    if (IS_945(intel->intelScreen->deviceID))
-      ok = i945_miptree_layout(intel, mt);
+      ok = i945_miptree_layout(intel, mt, tiling);
    else
-      ok = i915_miptree_layout(intel, mt);
+      ok = i915_miptree_layout(intel, mt, tiling);
 #else
-   ok = brw_miptree_layout(intel, mt);
+   ok = brw_miptree_layout(intel, mt, tiling);
 #endif
 
    if (!ok) {
@@ -109,10 +110,18 @@ intel_miptree_create(struct intel_context *intel,
 		     GLboolean expect_accelerated_upload)
 {
    struct intel_mipmap_tree *mt;
+   uint32_t tiling;
+
+   if (intel->use_texture_tiling && compress_byte == 0 &&
+       intel->intelScreen->kernel_exec_fencing)
+      tiling = I915_TILING_X;
+   else
+      tiling = I915_TILING_NONE;
 
    mt = intel_miptree_create_internal(intel, target, internal_format,
 				      first_level, last_level, width0,
-				      height0, depth0, cpp, compress_byte);
+				      height0, depth0, cpp, compress_byte,
+				      tiling);
    /*
     * pitch == 0 || height == 0  indicates the null texture
     */
@@ -120,6 +129,7 @@ intel_miptree_create(struct intel_context *intel,
       return NULL;
 
    mt->region = intel_region_alloc(intel,
+				   tiling,
 				   mt->cpp,
 				   mt->pitch,
 				   mt->total_height,
@@ -149,7 +159,8 @@ intel_miptree_create_for_region(struct intel_context *intel,
    mt = intel_miptree_create_internal(intel, target, internal_format,
 				      first_level, last_level,
 				      region->width, region->height, 1,
-				      region->cpp, compress_byte);
+				      region->cpp, compress_byte,
+				      I915_TILING_NONE);
    if (!mt)
       return mt;
 #if 0
@@ -187,6 +198,7 @@ intel_miptree_create_for_region(struct intel_context *intel,
 
 int intel_miptree_pitch_align (struct intel_context *intel,
 			       struct intel_mipmap_tree *mt,
+			       uint32_t tiling,
 			       int pitch)
 {
 #ifdef I915
@@ -207,6 +219,11 @@ int intel_miptree_pitch_align (struct intel_context *intel,
 	 pitch_align = 4;
       }
 
+      if (tiling == I915_TILING_X)
+	 pitch_align = 512;
+      else if (tiling == I915_TILING_Y)
+	 pitch_align = 128;
+
       pitch = ALIGN(pitch * mt->cpp, pitch_align);
 
 #ifdef I915
diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h
index 4060b9df78f..3af9966827f 100644
--- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h
+++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h
@@ -148,6 +148,7 @@ intel_miptree_create_for_region(struct intel_context *intel,
 
 int intel_miptree_pitch_align (struct intel_context *intel,
 			       struct intel_mipmap_tree *mt,
+			       uint32_t tiling,
 			       int pitch);
 
 void intel_miptree_reference(struct intel_mipmap_tree **dst,
@@ -218,10 +219,13 @@ void intel_miptree_image_copy(struct intel_context *intel,
 /* i915_mipmap_tree.c:
  */
 GLboolean i915_miptree_layout(struct intel_context *intel,
-			      struct intel_mipmap_tree *mt);
+			      struct intel_mipmap_tree *mt,
+			      uint32_t tiling);
 GLboolean i945_miptree_layout(struct intel_context *intel,
-			      struct intel_mipmap_tree *mt);
+			      struct intel_mipmap_tree *mt,
+			      uint32_t tiling);
 GLboolean brw_miptree_layout(struct intel_context *intel,
-			     struct intel_mipmap_tree *mt);
+			     struct intel_mipmap_tree *mt,
+			     uint32_t tiling);
 
 #endif
diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c
index 534e75efe1f..fd9bf7b1741 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.c
+++ b/src/mesa/drivers/dri/intel/intel_regions.c
@@ -116,7 +116,10 @@ intel_region_map(struct intel_context *intel, struct intel_region *region)
       if (region->pbo)
          intel_region_cow(intel, region);
 
-      dri_bo_map(region->buffer, GL_TRUE);
+      if (intel->intelScreen->kernel_exec_fencing)
+	 drm_intel_gem_bo_map_gtt(region->buffer);
+      else
+	 dri_bo_map(region->buffer, GL_TRUE);
       region->map = region->buffer->virtual;
    }
 
@@ -128,7 +131,10 @@ intel_region_unmap(struct intel_context *intel, struct intel_region *region)
 {
    _DBG("%s %p\n", __FUNCTION__, region);
    if (!--region->map_refcount) {
-      dri_bo_unmap(region->buffer);
+      if (intel->intelScreen->kernel_exec_fencing)
+	 drm_intel_gem_bo_map_gtt(region->buffer);
+      else
+	 dri_bo_unmap(region->buffer);
       region->map = NULL;
    }
 }
@@ -164,10 +170,12 @@ intel_region_alloc_internal(struct intel_context *intel,
 
 struct intel_region *
 intel_region_alloc(struct intel_context *intel,
+		   uint32_t tiling,
                    GLuint cpp, GLuint width, GLuint height, GLuint pitch,
 		   GLboolean expect_accelerated_upload)
 {
    dri_bo *buffer;
+   struct intel_region *region;
 
    if (expect_accelerated_upload) {
       buffer = drm_intel_bo_alloc_for_render(intel->bufmgr, "region",
@@ -177,7 +185,16 @@ intel_region_alloc(struct intel_context *intel,
 				  pitch * cpp * height, 64);
    }
 
-   return intel_region_alloc_internal(intel, cpp, width, height, pitch, buffer);
+   region = intel_region_alloc_internal(intel, cpp, width, height,
+					pitch, buffer);
+
+   if (tiling != I915_TILING_NONE) {
+      assert(((pitch * cpp) & 511) == 0);
+      drm_intel_bo_set_tiling(buffer, &tiling, pitch * cpp);
+      drm_intel_bo_get_tiling(buffer, &region->tiling, &region->bit_6_swizzle);
+   }
+
+   return region;
 }
 
 struct intel_region *
diff --git a/src/mesa/drivers/dri/intel/intel_regions.h b/src/mesa/drivers/dri/intel/intel_regions.h
index 45e2bf4e77a..bd3c8e7325b 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.h
+++ b/src/mesa/drivers/dri/intel/intel_regions.h
@@ -73,7 +73,8 @@ struct intel_region
  * copied by calling intel_reference_region().
  */
 struct intel_region *intel_region_alloc(struct intel_context *intel,
-                                        GLuint cpp, GLuint width,
+                                        uint32_t tiling,
+					GLuint cpp, GLuint width,
                                         GLuint height, GLuint pitch,
 					GLboolean expect_accelerated_upload);
 
diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c
index 27288231426..6521b4ef313 100644
--- a/src/mesa/drivers/dri/intel/intel_screen.c
+++ b/src/mesa/drivers/dri/intel/intel_screen.c
@@ -49,6 +49,10 @@
 #include "i915_drm.h"
 #include "i830_dri.h"
 
+#define DRI_CONF_TEXTURE_TILING(def) \
+	DRI_CONF_OPT_BEGIN(texture_tiling, bool, def)		\
+		DRI_CONF_DESC(en, "Enable texture tiling")	\
+	DRI_CONF_OPT_END					\
 
 PUBLIC const char __driConfigOptions[] =
    DRI_CONF_BEGIN
@@ -64,6 +68,13 @@ PUBLIC const char __driConfigOptions[] =
 	    DRI_CONF_ENUM(1, "Enable reuse of all sizes of buffer objects")
 	 DRI_CONF_DESC_END
       DRI_CONF_OPT_END
+
+#ifdef I915
+     DRI_CONF_TEXTURE_TILING(false)
+#else
+     DRI_CONF_TEXTURE_TILING(true)
+#endif
+
    DRI_CONF_SECTION_END
    DRI_CONF_SECTION_QUALITY
       DRI_CONF_FORCE_S3TC_ENABLE(false)
@@ -76,7 +87,7 @@ PUBLIC const char __driConfigOptions[] =
    DRI_CONF_SECTION_END
 DRI_CONF_END;
 
-const GLuint __driNConfigOptions = 8;
+const GLuint __driNConfigOptions = 9;
 
 #ifdef USE_NEW_INTERFACE
 static PFNGLXCREATECONTEXTMODES create_context_modes = NULL;
diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c
index a25626ae28a..673b8fa6a11 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_copy.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c
@@ -118,8 +118,12 @@ do_copy_texsubimage(struct intel_context *intel,
       dstx += x - orig_x;
       dsty += y - orig_y;
 
-      /* image_offset may be non-page-aligned, but that's illegal for tiling. */
-      assert(intelImage->mt->region->tiling == I915_TILING_NONE);
+      /* Can't blit to tiled buffers with non-tile-aligned offset. */
+      if (intelImage->mt->region->tiling != I915_TILING_NONE &&
+	  (image_offset & 4095) != 0) {
+	 UNLOCK_HARDWARE(intel);
+	 return GL_FALSE;
+      }
 
       if (ctx->ReadBuffer->Name == 0) {
 	 /* reading from a window, adjust x, y */
diff --git a/src/mesa/drivers/dri/intel/intel_tex_layout.c b/src/mesa/drivers/dri/intel/intel_tex_layout.c
index e6f9a417790..b8be7ef41a2 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_layout.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_layout.c
@@ -52,7 +52,9 @@ GLuint intel_compressed_alignment(GLenum internalFormat)
     return alignment;
 }
 
-void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tree *mt )
+void i945_miptree_layout_2d( struct intel_context *intel,
+			     struct intel_mipmap_tree *mt,
+			     uint32_t tiling )
 {
    GLint align_h = 2, align_w = 4;
    GLuint level;
@@ -86,13 +88,18 @@ void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tr
 
        if (mip1_width > mt->pitch) {
            mt->pitch = mip1_width;
+
+	   if (tiling == I915_TILING_X)
+	      mt->pitch = ALIGN(mt->pitch * mt->cpp, 512) / mt->cpp;
+	   if (tiling == I915_TILING_Y)
+	      mt->pitch = ALIGN(mt->pitch * mt->cpp, 128) / mt->cpp;
        }
    }
 
    /* Pitch must be a whole number of dwords, even though we
     * express it in texels.
     */
-   mt->pitch = intel_miptree_pitch_align (intel, mt, mt->pitch);
+   mt->pitch = intel_miptree_pitch_align (intel, mt, tiling, mt->pitch);
    mt->total_height = 0;
 
    for ( level = mt->first_level ; level <= mt->last_level ; level++ ) {
diff --git a/src/mesa/drivers/dri/intel/intel_tex_layout.h b/src/mesa/drivers/dri/intel/intel_tex_layout.h
index dbc90e6f9b7..7bc25b6bcb1 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_layout.h
+++ b/src/mesa/drivers/dri/intel/intel_tex_layout.h
@@ -38,5 +38,7 @@ static GLuint minify( GLuint d )
    return MAX2(1, d>>1);
 }
 
-extern void i945_miptree_layout_2d( struct intel_context *intel, struct intel_mipmap_tree *mt );
+extern void i945_miptree_layout_2d(struct intel_context *intel,
+				   struct intel_mipmap_tree *mt,
+				   uint32_t tiling);
 extern GLuint intel_compressed_alignment(GLenum);