From 6a49473ab5797b1e6ce021e396902f9cb77674ef Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Mon, 22 Jun 2009 10:54:50 -0700
Subject: intel: Remove long-unused intel_region_fill and intelEmitFillBlit.

---
 src/mesa/drivers/dri/intel/intel_blit.h | 9 ---------
 1 file changed, 9 deletions(-)

(limited to 'src/mesa/drivers/dri/intel/intel_blit.h')

diff --git a/src/mesa/drivers/dri/intel/intel_blit.h b/src/mesa/drivers/dri/intel/intel_blit.h
index 52065b13ed..2b4267ef6a 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.h
+++ b/src/mesa/drivers/dri/intel/intel_blit.h
@@ -50,15 +50,6 @@ extern void intelEmitCopyBlit(struct intel_context *intel,
                               GLshort w, GLshort h,
 			      GLenum logicop );
 
-extern void intelEmitFillBlit(struct intel_context *intel,
-                              GLuint cpp,
-                              GLshort dst_pitch,
-                              dri_bo *dst_buffer,
-                              GLuint dst_offset,
-			      uint32_t dst_tiling,
-                              GLshort x, GLshort y,
-                              GLshort w, GLshort h, GLuint color);
-
 void
 intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 				  GLuint cpp,
-- 
cgit v1.2.3


From 8f81a6468fdbc7320800ea497791e3e1b8f782ca Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Mon, 22 Jun 2009 11:00:11 -0700
Subject: intel: Avoid trying to do blits to Y tiled regions.

This is somewhat nasty, but we need to do Y-tiled depth for FBO support.
May help with corruption and hangs since enabling texture tiling, and
since switching depth textures to Y tiled.

Fixes piglit depthtex.c on 965.
---
 src/mesa/drivers/dri/intel/intel_blit.c         | 26 +++++++++++++----
 src/mesa/drivers/dri/intel/intel_blit.h         |  5 ++--
 src/mesa/drivers/dri/intel/intel_mipmap_tree.c  | 29 +++++++++++++-----
 src/mesa/drivers/dri/intel/intel_pixel_bitmap.c | 39 +++++++++++++------------
 src/mesa/drivers/dri/intel/intel_pixel_copy.c   | 16 ++++++----
 src/mesa/drivers/dri/intel/intel_pixel_read.c   | 23 ++++++++-------
 src/mesa/drivers/dri/intel/intel_regions.c      | 30 +++++++++----------
 src/mesa/drivers/dri/intel/intel_regions.h      | 17 ++++++-----
 src/mesa/drivers/dri/intel/intel_tex_copy.c     | 27 +++++++++--------
 src/mesa/drivers/dri/intel/intel_tex_image.c    | 15 ++++++----
 10 files changed, 137 insertions(+), 90 deletions(-)

(limited to 'src/mesa/drivers/dri/intel/intel_blit.h')

diff --git a/src/mesa/drivers/dri/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c
index 698a0bf465..e0596a5531 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.c
+++ b/src/mesa/drivers/dri/intel/intel_blit.c
@@ -108,6 +108,8 @@ intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
 	 CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
       }
 
+      assert(src->tiling != I915_TILING_Y);
+      assert(dst->tiling != I915_TILING_Y);
 #ifndef I915
       if (src->tiling != I915_TILING_NONE) {
 	 CMD |= XY_SRC_TILED;
@@ -201,7 +203,7 @@ static GLuint translate_raster_op(GLenum logicop)
 
 /* Copy BitBlt
  */
-void
+GLboolean
 intelEmitCopyBlit(struct intel_context *intel,
 		  GLuint cpp,
 		  GLshort src_pitch,
@@ -223,6 +225,11 @@ intelEmitCopyBlit(struct intel_context *intel,
    dri_bo *aper_array[3];
    BATCH_LOCALS;
 
+   if (dst_tiling == I915_TILING_Y)
+      return GL_FALSE;
+   if (src_tiling == I915_TILING_Y)
+      return GL_FALSE;
+
    /* do space/cliprects check before going any further */
    do {
        aper_array[0] = intel->batch->buf;
@@ -260,7 +267,7 @@ intelEmitCopyBlit(struct intel_context *intel,
        if (locked)
            UNLOCK_HARDWARE(intel);
 
-       return;
+       return GL_TRUE;
    }
 
    intel_batchbuffer_require_space(intel->batch, 8 * 4, NO_LOOP_CLIPRECTS);
@@ -287,7 +294,7 @@ intelEmitCopyBlit(struct intel_context *intel,
       CMD = XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
       break;
    default:
-      return;
+      return GL_FALSE;
    }
 
 #ifndef I915
@@ -302,7 +309,7 @@ intelEmitCopyBlit(struct intel_context *intel,
 #endif
 
    if (dst_y2 <= dst_y || dst_x2 <= dst_x) {
-      return;
+      return GL_TRUE;
    }
 
    assert(dst_x < dst_x2);
@@ -324,6 +331,8 @@ intelEmitCopyBlit(struct intel_context *intel,
    ADVANCE_BATCH();
 
    intel_batchbuffer_emit_mi_flush(intel->batch);
+
+   return GL_TRUE;
 }
 
 
@@ -536,7 +545,7 @@ intelClearWithBlit(GLcontext *ctx, GLbitfield mask)
    UNLOCK_HARDWARE(intel);
 }
 
-void
+GLboolean
 intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 				  GLuint cpp,
 				  GLubyte *src_bits, GLuint src_size,
@@ -552,11 +561,14 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
    int dwords = ALIGN(src_size, 8) / 4;
    uint32_t opcode, br13, blit_cmd;
 
+   if (dst_tiling == I915_TILING_Y)
+      return GL_FALSE;
+
    assert( logic_op - GL_CLEAR >= 0 );
    assert( logic_op - GL_CLEAR < 0x10 );
 
    if (w < 0 || h < 0)
-      return;
+      return GL_TRUE;
 
    dst_pitch *= cpp;
 
@@ -613,4 +625,6 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 			   REFERENCES_CLIPRECTS );
 
    intel_batchbuffer_emit_mi_flush(intel->batch);
+
+   return GL_TRUE;
 }
diff --git a/src/mesa/drivers/dri/intel/intel_blit.h b/src/mesa/drivers/dri/intel/intel_blit.h
index 2b4267ef6a..152fa3f17b 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.h
+++ b/src/mesa/drivers/dri/intel/intel_blit.h
@@ -35,7 +35,8 @@ extern void intelCopyBuffer(const __DRIdrawablePrivate * dpriv,
 
 extern void intelClearWithBlit(GLcontext * ctx, GLbitfield mask);
 
-extern void intelEmitCopyBlit(struct intel_context *intel,
+GLboolean
+intelEmitCopyBlit(struct intel_context *intel,
                               GLuint cpp,
                               GLshort src_pitch,
                               dri_bo *src_buffer,
@@ -50,7 +51,7 @@ extern void intelEmitCopyBlit(struct intel_context *intel,
                               GLshort w, GLshort h,
 			      GLenum logicop );
 
-void
+GLboolean
 intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 				  GLuint cpp,
 				  GLubyte *src_bits, GLuint src_size,
diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
index fadcc89b3e..660d7e5350 100644
--- a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
+++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
@@ -498,6 +498,7 @@ intel_miptree_image_copy(struct intel_context *intel,
    const GLuint *dst_depth_offset = intel_miptree_depth_offsets(dst, level);
    const GLuint *src_depth_offset = intel_miptree_depth_offsets(src, level);
    GLuint i;
+   GLboolean success;
 
    if (dst->compressed) {
        GLuint alignment = intel_compressed_alignment(dst->internal_format);
@@ -506,12 +507,26 @@ intel_miptree_image_copy(struct intel_context *intel,
    }
 
    for (i = 0; i < depth; i++) {
-      intel_region_copy(intel,
-                        dst->region, dst_offset + dst_depth_offset[i],
-                        0,
-                        0,
-                        src->region, src_offset + src_depth_offset[i],
-                        0, 0, width, height, GL_COPY);
+      success = intel_region_copy(intel,
+				  dst->region, dst_offset + dst_depth_offset[i],
+				  0, 0,
+				  src->region, src_offset + src_depth_offset[i],
+				  0, 0, width, height, GL_COPY);
+      if (!success) {
+	 GLubyte *src_ptr, *dst_ptr;
+
+	 src_ptr = intel_region_map(intel, src->region);
+	 dst_ptr = intel_region_map(intel, dst->region);
+
+	 _mesa_copy_rect(dst_ptr + dst_offset + dst_depth_offset[i],
+			 dst->cpp,
+			 dst->pitch,
+			 0, 0, width, height,
+			 src_ptr + src_offset + src_depth_offset[i],
+			 src->pitch,
+			 0, 0);
+	 intel_region_unmap(intel, src->region);
+	 intel_region_unmap(intel, dst->region);
+      }
    }
-
 }
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
index 191d9b0846..e678cd2c26 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_bitmap.c
@@ -194,7 +194,7 @@ do_blit_bitmap( GLcontext *ctx,
    struct gl_framebuffer *fb = ctx->DrawBuffer;
    GLfloat tmpColor[4];
    GLubyte ubcolor[4];
-   GLuint color8888, color565;
+   GLuint color;
    unsigned int num_cliprects;
    drm_clip_rect_t *cliprects;
    int x_off, y_off;
@@ -232,8 +232,11 @@ do_blit_bitmap( GLcontext *ctx,
    UNCLAMPED_FLOAT_TO_UBYTE(ubcolor[2], tmpColor[2]);
    UNCLAMPED_FLOAT_TO_UBYTE(ubcolor[3], tmpColor[3]);
 
-   color8888 = INTEL_PACKCOLOR8888(ubcolor[0], ubcolor[1], ubcolor[2], ubcolor[3]);
-   color565 = INTEL_PACKCOLOR565(ubcolor[0], ubcolor[1], ubcolor[2]);
+   if (dst->cpp == 2)
+      color = INTEL_PACKCOLOR565(ubcolor[0], ubcolor[1], ubcolor[2]);
+   else
+      color = INTEL_PACKCOLOR8888(ubcolor[0], ubcolor[1],
+				  ubcolor[2], ubcolor[3]);
 
    if (!intel_check_blit_fragment_ops(ctx, tmpColor[3] == 1.0F))
       return GL_FALSE;
@@ -307,21 +310,21 @@ do_blit_bitmap( GLcontext *ctx,
 				   fb->Name == 0 ? GL_TRUE : GL_FALSE) == 0)
 		  continue;
 
-	       /* 
-		*/
-	       intelEmitImmediateColorExpandBlit( intel,
-						  dst->cpp,
-						  (GLubyte *)stipple, 
-						  sz,
-						  (dst->cpp == 2) ? color565 : color8888,
-						  dst->pitch,
-						  dst->buffer,
-						  0,
-						  dst->tiling,
-						  box_x + px,
-						  box_y + py,
-						  w, h,
-						  logic_op);
+	       if (!intelEmitImmediateColorExpandBlit(intel,
+						      dst->cpp,
+						      (GLubyte *)stipple,
+						      sz,
+						      color,
+						      dst->pitch,
+						      dst->buffer,
+						      0,
+						      dst->tiling,
+						      box_x + px,
+						      box_y + py,
+						      w, h,
+						      logic_op)) {
+		  return GL_FALSE;
+	       }
 	    } 
 	 } 
       }
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_copy.c b/src/mesa/drivers/dri/intel/intel_pixel_copy.c
index 1a3e363846..f523d3eead 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_copy.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_copy.c
@@ -362,12 +362,16 @@ do_blit_copypixels(GLcontext * ctx,
 				   &clip_x, &clip_y, &clip_w, &clip_h))
             continue;
 
-	 intel_region_copy(intel,
-			   dst, 0, clip_x, clip_y,
-			   src, 0, clip_x + delta_x, clip_y + delta_y,
-			   clip_w, clip_h,
-			   ctx->Color.ColorLogicOpEnabled ?
-			   ctx->Color.LogicOp : GL_COPY);
+	 if (!intel_region_copy(intel,
+				dst, 0, clip_x, clip_y,
+				src, 0, clip_x + delta_x, clip_y + delta_y,
+				clip_w, clip_h,
+				ctx->Color.ColorLogicOpEnabled ?
+				ctx->Color.LogicOp : GL_COPY)) {
+	    DBG("%s: blit failure\n", __FUNCTION__);
+	    UNLOCK_HARDWARE(intel);
+	    return GL_FALSE;
+	 }
       }
    }
 out:
diff --git a/src/mesa/drivers/dri/intel/intel_pixel_read.c b/src/mesa/drivers/dri/intel/intel_pixel_read.c
index 74da6d39a2..8713463ace 100644
--- a/src/mesa/drivers/dri/intel/intel_pixel_read.c
+++ b/src/mesa/drivers/dri/intel/intel_pixel_read.c
@@ -261,16 +261,19 @@ do_blit_readpixels(GLcontext * ctx,
          if (!intel_intersect_cliprects(&rect, &src_rect, &box[i]))
             continue;
 
-         intelEmitCopyBlit(intel,
-                           src->cpp,
-                           src->pitch, src->buffer, 0, src->tiling,
-                           rowLength, dst_buffer, dst_offset, GL_FALSE,
-                           rect.x1,
-                           rect.y1,
-                           rect.x1 - src_rect.x1,
-                           rect.y2 - src_rect.y2,
-                           rect.x2 - rect.x1, rect.y2 - rect.y1,
-			   GL_COPY);
+         if (!intelEmitCopyBlit(intel,
+				src->cpp,
+				src->pitch, src->buffer, 0, src->tiling,
+				rowLength, dst_buffer, dst_offset, GL_FALSE,
+				rect.x1,
+				rect.y1,
+				rect.x1 - src_rect.x1,
+				rect.y2 - src_rect.y2,
+				rect.x2 - rect.x1, rect.y2 - rect.y1,
+				GL_COPY)) {
+	    UNLOCK_HARDWARE(intel);
+	    return GL_FALSE;
+	 }
       }
    }
    UNLOCK_HARDWARE(intel);
diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c
index 98109235b7..4ce7f12e40 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.c
+++ b/src/mesa/drivers/dri/intel/intel_regions.c
@@ -360,7 +360,7 @@ intel_region_data(struct intel_context *intel,
 /* Copy rectangular sub-regions. Need better logic about when to
  * push buffers into AGP - will currently do so whenever possible.
  */
-void
+GLboolean
 intel_region_copy(struct intel_context *intel,
                   struct intel_region *dst,
                   GLuint dst_offset,
@@ -373,7 +373,7 @@ intel_region_copy(struct intel_context *intel,
    _DBG("%s\n", __FUNCTION__);
 
    if (intel == NULL)
-      return;
+      return GL_FALSE;
 
    if (dst->pbo) {
       if (dstx == 0 &&
@@ -385,12 +385,12 @@ intel_region_copy(struct intel_context *intel,
 
    assert(src->cpp == dst->cpp);
 
-   intelEmitCopyBlit(intel,
-                     dst->cpp,
-                     src->pitch, src->buffer, src_offset, src->tiling,
-                     dst->pitch, dst->buffer, dst_offset, dst->tiling,
-                     srcx, srcy, dstx, dsty, width, height,
-		     logicop);
+   return intelEmitCopyBlit(intel,
+			    dst->cpp,
+			    src->pitch, src->buffer, src_offset, src->tiling,
+			    dst->pitch, dst->buffer, dst_offset, dst->tiling,
+			    srcx, srcy, dstx, dsty, width, height,
+			    logicop);
 }
 
 /* Attach to a pbo, discarding our data.  Effectively zero-copy upload
@@ -477,13 +477,13 @@ intel_region_cow(struct intel_context *intel, struct intel_region *region)
    if (!was_locked)
       LOCK_HARDWARE(intel);
 
-   intelEmitCopyBlit(intel,
-		     region->cpp,
-		     region->pitch, pbo->buffer, 0, region->tiling,
-		     region->pitch, region->buffer, 0, region->tiling,
-		     0, 0, 0, 0,
-		     region->pitch, region->height,
-		     GL_COPY);
+   assert(intelEmitCopyBlit(intel,
+			    region->cpp,
+			    region->pitch, pbo->buffer, 0, region->tiling,
+			    region->pitch, region->buffer, 0, region->tiling,
+			    0, 0, 0, 0,
+			    region->pitch, region->height,
+			    GL_COPY));
 
    if (!was_locked)
       UNLOCK_HARDWARE(intel);
diff --git a/src/mesa/drivers/dri/intel/intel_regions.h b/src/mesa/drivers/dri/intel/intel_regions.h
index c2d4810df2..0d379bdc6e 100644
--- a/src/mesa/drivers/dri/intel/intel_regions.h
+++ b/src/mesa/drivers/dri/intel/intel_regions.h
@@ -110,14 +110,15 @@ void intel_region_data(struct intel_context *intel,
 
 /* Copy rectangular sub-regions
  */
-void intel_region_copy(struct intel_context *intel,
-                       struct intel_region *dest,
-                       GLuint dest_offset,
-                       GLuint destx, GLuint desty,
-                       struct intel_region *src,
-                       GLuint src_offset,
-                       GLuint srcx, GLuint srcy, GLuint width, GLuint height,
-		       GLenum logicop);
+GLboolean
+intel_region_copy(struct intel_context *intel,
+		  struct intel_region *dest,
+		  GLuint dest_offset,
+		  GLuint destx, GLuint desty,
+		  struct intel_region *src,
+		  GLuint src_offset,
+		  GLuint srcx, GLuint srcy, GLuint width, GLuint height,
+		  GLenum logicop);
 
 /* Helpers for zerocopy uploads, particularly texture image uploads:
  */
diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c
index e1820e3fcf..0335c13307 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_copy.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c
@@ -147,18 +147,21 @@ do_copy_texsubimage(struct intel_context *intel,
 	 src_pitch = src->pitch;
       }
 
-      intelEmitCopyBlit(intel,
-			intelImage->mt->cpp,
-			src_pitch,
-			src->buffer,
-			0,
-			src->tiling,
-			intelImage->mt->pitch,
-			dst_bo,
-			image_offset,
-			intelImage->mt->region->tiling,
-			x, y, dstx, dsty, width, height,
-			GL_COPY);
+      if (!intelEmitCopyBlit(intel,
+			     intelImage->mt->cpp,
+			     src_pitch,
+			     src->buffer,
+			     0,
+			     src->tiling,
+			     intelImage->mt->pitch,
+			     dst_bo,
+			     image_offset,
+			     intelImage->mt->region->tiling,
+			     x, y, dstx, dsty, width, height,
+			     GL_COPY)) {
+	 UNLOCK_HARDWARE(intel);
+	 return GL_FALSE;
+      }
    }
 
    UNLOCK_HARDWARE(intel);
diff --git a/src/mesa/drivers/dri/intel/intel_tex_image.c b/src/mesa/drivers/dri/intel/intel_tex_image.c
index 03311e58aa..ab9541359d 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_image.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_image.c
@@ -235,12 +235,15 @@ try_pbo_upload(struct intel_context *intel,
 					       INTEL_WRITE_FULL);
 
 
-      intelEmitCopyBlit(intel,
-                        intelImage->mt->cpp,
-                        src_stride, src_buffer, src_offset, GL_FALSE,
-                        dst_stride, dst_buffer, dst_offset, GL_FALSE,
-                        0, 0, 0, 0, width, height,
-			GL_COPY);
+      if (!intelEmitCopyBlit(intel,
+			     intelImage->mt->cpp,
+			     src_stride, src_buffer, src_offset, GL_FALSE,
+			     dst_stride, dst_buffer, dst_offset, GL_FALSE,
+			     0, 0, 0, 0, width, height,
+			     GL_COPY)) {
+	 UNLOCK_HARDWARE(intel);
+	 return GL_FALSE;
+      }
    }
    UNLOCK_HARDWARE(intel);
 
-- 
cgit v1.2.3


From 2d5c74fac3cc0f9d45a9e11b2fcdea1bc67928c4 Mon Sep 17 00:00:00 2001
From: Eric Anholt <eric@anholt.net>
Date: Wed, 26 Aug 2009 18:08:52 -0700
Subject: intel: Add support for GL_ARB_map_buffer_range.

Passes glean's bufferObject test, and should provide good performance in the
cases applications are expected to use.
---
 src/mesa/drivers/dri/intel/intel_blit.c           |  40 +++++
 src/mesa/drivers/dri/intel/intel_blit.h           |   6 +
 src/mesa/drivers/dri/intel/intel_buffer_objects.c | 171 +++++++++++++++++-----
 src/mesa/drivers/dri/intel/intel_buffer_objects.h |   5 +
 src/mesa/drivers/dri/intel/intel_extensions.c     |   2 +
 5 files changed, 189 insertions(+), 35 deletions(-)

(limited to 'src/mesa/drivers/dri/intel/intel_blit.h')

diff --git a/src/mesa/drivers/dri/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c
index 979f202584..0c5be4c798 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.c
+++ b/src/mesa/drivers/dri/intel/intel_blit.c
@@ -636,3 +636,43 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 
    return GL_TRUE;
 }
+
+/* We don't have a memmove-type blit like some other hardware, so we'll do a
+ * rectangular blit covering a large space, then emit 1-scanline blit at the
+ * end to cover the last if we need.
+ */
+void
+intel_emit_linear_blit(struct intel_context *intel,
+		       drm_intel_bo *dst_bo,
+		       unsigned int dst_offset,
+		       drm_intel_bo *src_bo,
+		       unsigned int src_offset,
+		       unsigned int size)
+{
+   GLuint pitch, height;
+
+   /* The pitch is a signed value. */
+   pitch = MIN2(size, (1 << 15) - 1);
+   height = size / pitch;
+   intelEmitCopyBlit(intel, 1,
+		     pitch, src_bo, src_offset, I915_TILING_NONE,
+		     pitch, dst_bo, dst_offset, I915_TILING_NONE,
+		     0, 0, /* src x/y */
+		     0, 0, /* dst x/y */
+		     pitch, height, /* w, h */
+		     GL_COPY);
+
+   src_offset += pitch * height;
+   dst_offset += pitch * height;
+   size -= pitch * height;
+   assert (size < (1 << 15));
+   if (size != 0) {
+      intelEmitCopyBlit(intel, 1,
+			size, src_bo, src_offset, I915_TILING_NONE,
+			size, dst_bo, dst_offset, I915_TILING_NONE,
+			0, 0, /* src x/y */
+			0, 0, /* dst x/y */
+			size, 1, /* w, h */
+			GL_COPY);
+   }
+}
diff --git a/src/mesa/drivers/dri/intel/intel_blit.h b/src/mesa/drivers/dri/intel/intel_blit.h
index 152fa3f17b..240cb7cd1b 100644
--- a/src/mesa/drivers/dri/intel/intel_blit.h
+++ b/src/mesa/drivers/dri/intel/intel_blit.h
@@ -63,5 +63,11 @@ intelEmitImmediateColorExpandBlit(struct intel_context *intel,
 				  GLshort x, GLshort y,
 				  GLshort w, GLshort h,
 				  GLenum logic_op);
+void intel_emit_linear_blit(struct intel_context *intel,
+			    drm_intel_bo *dst_bo,
+			    unsigned int dst_offset,
+			    drm_intel_bo *src_bo,
+			    unsigned int src_offset,
+			    unsigned int size);
 
 #endif
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.c b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
index a740397943..3cf0ea0991 100644
--- a/src/mesa/drivers/dri/intel/intel_buffer_objects.c
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
@@ -270,29 +270,155 @@ intel_bufferobj_map(GLcontext * ctx,
    return obj->Pointer;
 }
 
+/**
+ * Called via glMapBufferRange().
+ *
+ * The goal of this extension is to allow apps to accumulate their rendering
+ * at the same time as they accumulate their buffer object.  Without it,
+ * you'd end up blocking on execution of rendering every time you mapped
+ * the buffer to put new data in.
+ *
+ * We support it in 3 ways: If unsynchronized, then don't bother
+ * flushing the batchbuffer before mapping the buffer, which can save blocking
+ * in many cases.  If we would still block, and they allow the whole buffer
+ * to be invalidated, then just allocate a new buffer to replace the old one.
+ * If not, and we'd block, and they allow the subrange of the buffer to be
+ * invalidated, then we can make a new little BO, let them write into that,
+ * and blit it into the real BO at unmap time.
+ */
+static void *
+intel_bufferobj_map_range(GLcontext * ctx,
+			  GLenum target, GLsizei offset, GLsizeiptr length,
+			  GLbitfield access, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+
+   if (intel_obj->sys_buffer) {
+      obj->Pointer = intel_obj->sys_buffer + offset;
+      return obj->Pointer;
+   }
+
+   if (intel_obj->region)
+      intel_bufferobj_cow(intel, intel_obj);
+
+   /* If the mapping is synchronized with other GL operations, flush
+    * the batchbuffer so that GEM knows about the buffer access for later
+    * syncing.
+    */
+   if ((access & GL_MAP_WRITE_BIT) && !(access & GL_MAP_UNSYNCHRONIZED_BIT))
+      intelFlush(ctx);
+
+   /* _mesa_MapBufferRange (GL entrypoint) sets these, but the vbo module also
+    * internally uses our functions directly.
+    */
+   obj->Offset = offset;
+   obj->Length = length;
+   obj->AccessFlags = access;
+
+   if (intel_obj->buffer == NULL) {
+      obj->Pointer = NULL;
+      return NULL;
+   }
+
+   /* If the user doesn't care about existing buffer contents and mapping
+    * would cause us to block, then throw out the old buffer.
+    */
+   if (!(access & GL_MAP_UNSYNCHRONIZED_BIT) &&
+       (access & GL_MAP_INVALIDATE_BUFFER_BIT) &&
+       drm_intel_bo_busy(intel_obj->buffer)) {
+      drm_intel_bo_unreference(intel_obj->buffer);
+      intel_obj->buffer = dri_bo_alloc(intel->bufmgr, "bufferobj",
+				       intel_obj->Base.Size, 64);
+   }
+
+   /* If the user is mapping a range of an active buffer object but
+    * doesn't require the current contents of that range, make a new
+    * BO, and we'll copy what they put in there out at unmap or
+    * FlushRange time.
+    */
+   if ((access & GL_MAP_INVALIDATE_RANGE_BIT) &&
+       drm_intel_bo_busy(intel_obj->buffer)) {
+      intel_obj->range_map_bo = drm_intel_bo_alloc(intel->bufmgr,
+						   "range map",
+						   length, 64);
+      if (!(access & GL_MAP_READ_BIT) &&
+	  intel->intelScreen->kernel_exec_fencing) {
+	 drm_intel_gem_bo_map_gtt(intel_obj->range_map_bo);
+	 intel_obj->mapped_gtt = GL_TRUE;
+      } else {
+	 drm_intel_bo_map(intel_obj->range_map_bo,
+			  (access & GL_MAP_WRITE_BIT) != 0);
+	 intel_obj->mapped_gtt = GL_FALSE;
+      }
+      obj->Pointer = intel_obj->range_map_bo->virtual;
+      return obj->Pointer;
+   }
+
+   if (!(access & GL_MAP_READ_BIT) &&
+       intel->intelScreen->kernel_exec_fencing) {
+      drm_intel_gem_bo_map_gtt(intel_obj->buffer);
+      intel_obj->mapped_gtt = GL_TRUE;
+   } else {
+      drm_intel_bo_map(intel_obj->buffer, (access & GL_MAP_WRITE_BIT) != 0);
+      intel_obj->mapped_gtt = GL_FALSE;
+   }
+
+   obj->Pointer = intel_obj->buffer->virtual + offset;
+   return obj->Pointer;
+}
+
 
 /**
- * Called via glMapBufferARB().
+ * Called via glUnmapBuffer().
  */
 static GLboolean
 intel_bufferobj_unmap(GLcontext * ctx,
                       GLenum target, struct gl_buffer_object *obj)
 {
+   struct intel_context *intel = intel_context(ctx);
    struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
 
    assert(intel_obj);
+   assert(obj->Pointer);
    if (intel_obj->sys_buffer != NULL) {
-      assert(obj->Pointer);
-      obj->Pointer = NULL;
+      /* always keep the mapping around. */
+   } else if (intel_obj->range_map_bo != NULL) {
+      if (intel_obj->mapped_gtt) {
+	 drm_intel_gem_bo_unmap_gtt(intel_obj->range_map_bo);
+      } else {
+	 drm_intel_bo_unmap(intel_obj->range_map_bo);
+      }
+
+      /* We ignore the FLUSH_EXPLICIT bit and the calls associated with it.
+       * It would be a small win to support that, but for now we just copy
+       * the whole mapped range into place.
+       */
+      intel_emit_linear_blit(intel,
+			     intel_obj->buffer, obj->Offset,
+			     intel_obj->range_map_bo, 0,
+			     obj->Length);
+
+      /* Since we've emitted some blits to buffers that will (likely) be used
+       * in rendering operations in other cache domains in this batch, emit a
+       * flush.  Once again, we wish for a domain tracker in libdrm to cover
+       * usage inside of a batchbuffer.
+       */
+      intel_batchbuffer_emit_mi_flush(intel->batch);
+
+      drm_intel_bo_unreference(intel_obj->range_map_bo);
+      intel_obj->range_map_bo = NULL;
    } else if (intel_obj->buffer != NULL) {
-      assert(obj->Pointer);
       if (intel_obj->mapped_gtt) {
 	 drm_intel_gem_bo_unmap_gtt(intel_obj->buffer);
       } else {
 	 drm_intel_bo_unmap(intel_obj->buffer);
       }
-      obj->Pointer = NULL;
    }
+   obj->Pointer = NULL;
+
    return GL_TRUE;
 }
 
@@ -340,7 +466,6 @@ intel_bufferobj_copy_subdata(GLcontext *ctx,
    struct intel_buffer_object *intel_src = intel_buffer_object(src);
    struct intel_buffer_object *intel_dst = intel_buffer_object(dst);
    drm_intel_bo *src_bo, *dst_bo;
-   GLuint pitch, height;
 
    if (size == 0)
       return;
@@ -371,39 +496,14 @@ intel_bufferobj_copy_subdata(GLcontext *ctx,
       }
    }
 
-   /* Otherwise, we have real BOs, so blit them.  We don't have a memmove-type
-    * blit like some other hardware, so we'll do a rectangular blit covering
-    * a large space, then emit a scanline blit at the end to cover the last
-    * if we need.
-    */
+   /* Otherwise, we have real BOs, so blit them. */
 
    dst_bo = intel_bufferobj_buffer(intel, intel_dst, INTEL_WRITE_PART);
    src_bo = intel_bufferobj_buffer(intel, intel_src, INTEL_READ);
 
-   /* The pitch is a signed value. */
-   pitch = MIN2(size, (1 << 15) - 1);
-   height = size / pitch;
-   intelEmitCopyBlit(intel, 1,
-		     pitch, src_bo, read_offset, I915_TILING_NONE,
-		     pitch, dst_bo, write_offset, I915_TILING_NONE,
-		     0, 0, /* src x/y */
-		     0, 0, /* dst x/y */
-		     pitch, height, /* w, h */
-		     GL_COPY);
-
-   read_offset += pitch * height;
-   write_offset += pitch * height;
-   size -= pitch * height;
-   assert (size < (1 << 15));
-   if (size != 0) {
-      intelEmitCopyBlit(intel, 1,
-			size, src_bo, read_offset, I915_TILING_NONE,
-			size, dst_bo, write_offset, I915_TILING_NONE,
-			0, 0, /* src x/y */
-			0, 0, /* dst x/y */
-			size, 1, /* w, h */
-			GL_COPY);
-   }
+   intel_emit_linear_blit(intel,
+			  dst_bo, write_offset,
+			  src_bo, read_offset, size);
 
    /* Since we've emitted some blits to buffers that will (likely) be used
     * in rendering operations in other cache domains in this batch, emit a
@@ -422,6 +522,7 @@ intelInitBufferObjectFuncs(struct dd_function_table *functions)
    functions->BufferSubData = intel_bufferobj_subdata;
    functions->GetBufferSubData = intel_bufferobj_get_subdata;
    functions->MapBuffer = intel_bufferobj_map;
+   functions->MapBufferRange = intel_bufferobj_map_range;
    functions->UnmapBuffer = intel_bufferobj_unmap;
    functions->CopyBufferSubData = intel_bufferobj_copy_subdata;
 }
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.h b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
index 8164407f07..06a8ab9824 100644
--- a/src/mesa/drivers/dri/intel/intel_buffer_objects.h
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
@@ -48,6 +48,11 @@ struct intel_buffer_object
    struct intel_region *region; /* Is there a zero-copy texture
                                    associated with this (pixel)
                                    buffer object? */
+
+   drm_intel_bo *range_map_bo;
+   unsigned int range_map_offset;
+   GLsizei range_map_size;
+
    GLboolean mapped_gtt;
 };
 
diff --git a/src/mesa/drivers/dri/intel/intel_extensions.c b/src/mesa/drivers/dri/intel/intel_extensions.c
index ff9ad5acce..7aee70f0a8 100644
--- a/src/mesa/drivers/dri/intel/intel_extensions.c
+++ b/src/mesa/drivers/dri/intel/intel_extensions.c
@@ -32,6 +32,7 @@
 
 #define need_GL_ARB_copy_buffer
 #define need_GL_ARB_framebuffer_object
+#define need_GL_ARB_map_buffer_range
 #define need_GL_ARB_occlusion_query
 #define need_GL_ARB_point_parameters
 #define need_GL_ARB_shader_objects
@@ -72,6 +73,7 @@
 static const struct dri_extension card_extensions[] = {
    { "GL_ARB_copy_buffer",                GL_ARB_copy_buffer_functions },
    { "GL_ARB_half_float_pixel",           NULL },
+   { "GL_ARB_map_buffer_range",           GL_ARB_map_buffer_range_functions },
    { "GL_ARB_multitexture",               NULL },
    { "GL_ARB_point_parameters",           GL_ARB_point_parameters_functions },
    { "GL_ARB_point_sprite",               NULL },
-- 
cgit v1.2.3