i965: Implement CopyTexSubImage2D via BLORP (and use it by default).

The BLT engine has many limitations.  Currently, it can only blit
X-tiled buffers (since we don't have a kernel API to whack the BLT
tiling mode register), which means all depth/stencil operations get
punted to meta code, which can be very CPU-intensive.

Even if we used the BLT engine, it can't blit between buffers with
different tiling modes, such as an X-tiled non-MSAA ARGB8888 texture
and a Y-tiled CMS ARGB8888 renderbuffer.  This is a fundamental
limitation, and the only way around that is to use BLORP.

Previously, BLORP only handled BlitFramebuffer.  This patch adds an
additional frontend for doing CopyTexSubImage.  It also makes it the
default.  This is partly to increase testing and avoid hiding bugs,
and partly because the BLORP path can already handle more cases.  With
trivial extensions, it should be able to handle everything the BLT can.

This helps PlaneShift massively, which tries to CopyTexSubImage2D
between depth buffers whenever a player casts a spell.  Since these
are Y-tiled, we hit meta and software ReadPixels paths, eating 99% CPU
while delivering ~1 FPS.  This is particularly bad in an MMO setting
because people cast spells all the time.

It also helps Xonotic in 4X MSAA mode.  At default power management
settings, I measured a 6.35138% +/- 0.672548% performance boost (n=5).
(This data is from v1 of the patch.)

No Piglit regressions on Ivybridge (v3) or Sandybridge (v2).

v2: Create a fake intel_renderbuffer to wrap the destination texture
    image and then reuse do_blorp_blit rather than reimplementing most
    of it.  Remove unnecessary clipping code and conditional rendering
    check.

v3: Reuse formats_match() to centralize checks; delete temporary
    renderbuffers.  Reorganize the code.

v4: Actually copy stencil when dealing with separate stencil buffers but
    packed depth/stencil formats.  Tested by a new Piglit test.

NOTE: This is a candidate for the 9.1 branch.

Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Paul Berry <[email protected]> [v4]
Reviewed-by: Ian Romanick <[email protected]> [v3]
Reviewed-and-tested-by: Carl Worth <[email protected]> [v2]
Tested-by: Martin Steigerwald <[email protected]> [v3]

5 files changed, 154 insertions, 8 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
index bc7916a1bd4..f45bbbf4315 100644
--- a/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
+++ b/src/mesa/drivers/dri/i965/brw_blorp_blit.cpp
@@ -23,6 +23,7 @@
 
 #include "main/teximage.h"
 #include "main/fbobject.h"
+#include "main/renderbuffer.h"
 
 #include "glsl/ralloc.h"
 
@@ -295,6 +296,93 @@ try_blorp_blit(struct intel_context *intel,
    return true;
 }
 
+bool
+brw_blorp_copytexsubimage(struct intel_context *intel,
+                          struct gl_renderbuffer *src_rb,
+                          struct gl_texture_image *dst_image,
+                          int srcX0, int srcY0,
+                          int dstX0, int dstY0,
+                          int width, int height)
+{
+   struct gl_context *ctx = &intel->ctx;
+   struct intel_renderbuffer *src_irb = intel_renderbuffer(src_rb);
+   struct intel_renderbuffer *dst_irb;
+
+   /* BLORP is not supported before Gen6. */
+   if (intel->gen < 6)
+      return false;
+
+   /* Create a fake/wrapper renderbuffer to allow us to use do_blorp_blit(). */
+   dst_irb = intel_create_fake_renderbuffer_wrapper(intel, dst_image);
+   if (!dst_irb)
+      return false;
+
+   struct gl_renderbuffer *dst_rb = &dst_irb->Base.Base;
+
+   /* Unlike BlitFramebuffer, CopyTexSubImage doesn't have a buffer bit.
+    * It's only used by find_miptee() to decide whether to dereference the
+    * separate stencil miptree.  In the case of packed depth/stencil, core
+    * Mesa hands us the depth attachment as src_rb (not stencil), so assume
+    * non-stencil for now.  A buffer bit of 0 works for both color and depth.
+    */
+   GLbitfield buffer_bit = 0;
+
+   if (!formats_match(buffer_bit, src_irb, dst_irb)) {
+      _mesa_delete_renderbuffer(ctx, dst_rb);
+      return false;
+   }
+
+   /* Source clipping shouldn't be necessary, since copytexsubimage (in
+    * src/mesa/main/teximage.c) calls _mesa_clip_copytexsubimage() which
+    * takes care of it.
+    *
+    * Destination clipping shouldn't be necessary since the restrictions on
+    * glCopyTexSubImage prevent the user from specifying a destination rectangle
+    * that falls outside the bounds of the destination texture.
+    * See error_check_subtexture_dimensions().
+    */
+
+   int srcY1 = srcY0 + height;
+   int dstX1 = dstX0 + width;
+   int dstY1 = dstY0 + height;
+
+   /* Sync up the state of window system buffers.  We need to do this before
+    * we go looking for the buffers.
+    */
+   intel_prepare_render(intel);
+
+   /* Account for the fact that in the system framebuffer, the origin is at
+    * the lower left.
+    */
+   bool mirror_y = false;
+   if (_mesa_is_winsys_fbo(ctx->ReadBuffer)) {
+      GLint tmp = src_rb->Height - srcY0;
+      srcY0 = src_rb->Height - srcY1;
+      srcY1 = tmp;
+      mirror_y = true;
+   }
+
+   do_blorp_blit(intel, buffer_bit, src_irb, dst_irb,
+                 srcX0, srcY0, dstX0, dstY0, dstX1, dstY1, false, mirror_y);
+
+   /* If we're copying a packed depth stencil texture, the above do_blorp_blit
+    * copied depth (since buffer_bit != GL_STENCIL_BIT).  Now copy stencil as
+    * well.  There's no need to do a formats_match() check because the separate
+    * stencil buffer is always S8.
+    */
+   src_rb = ctx->ReadBuffer->Attachment[BUFFER_STENCIL].Renderbuffer;
+   if (_mesa_get_format_bits(dst_image->TexFormat, GL_STENCIL_BITS) > 0 &&
+       src_rb != NULL) {
+      src_irb = intel_renderbuffer(src_rb);
+      do_blorp_blit(intel, GL_STENCIL_BUFFER_BIT, src_irb, dst_irb,
+                    srcX0, srcY0, dstX0, dstY0, dstX1, dstY1, false, mirror_y);
+   }
+
+   _mesa_delete_renderbuffer(ctx, dst_rb);
+   return true;
+}
+
+
 GLbitfield
 brw_blorp_framebuffer(struct intel_context *intel,
                       GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 9affb7e5674..50956029590 100644
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -1215,6 +1215,14 @@ brw_blorp_framebuffer(struct intel_context *intel,
                       GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
                       GLbitfield mask, GLenum filter);
 
+bool
+brw_blorp_copytexsubimage(struct intel_context *intel,
+                          struct gl_renderbuffer *src_rb,
+                          struct gl_texture_image *dst_image,
+                          int srcX0, int srcY0,
+                          int dstX0, int dstY0,
+                          int width, int height);
+
 /* gen6_multisample_state.c */
 void
 gen6_emit_3dstate_multisample(struct brw_context *brw,
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c
index 481080944dd..37ecbd1701d 100644
--- a/src/mesa/drivers/dri/intel/intel_fbo.c
+++ b/src/mesa/drivers/dri/intel/intel_fbo.c
@@ -531,6 +531,36 @@ intel_renderbuffer_update_wrapper(struct intel_context *intel,
    return true;
 }
 
+/**
+ * Create a fake intel_renderbuffer that wraps a gl_texture_image.
+ */
+struct intel_renderbuffer *
+intel_create_fake_renderbuffer_wrapper(struct intel_context *intel,
+                                       struct gl_texture_image *image)
+{
+   struct gl_context *ctx = &intel->ctx;
+   struct intel_renderbuffer *irb;
+   struct gl_renderbuffer *rb;
+
+   irb = CALLOC_STRUCT(intel_renderbuffer);
+   if (!irb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "creating renderbuffer");
+      return NULL;
+   }
+
+   rb = &irb->Base.Base;
+
+   _mesa_init_renderbuffer(rb, 0);
+   rb->ClassID = INTEL_RB_CLASS;
+
+   if (!intel_renderbuffer_update_wrapper(intel, irb, image, image->Face)) {
+      intel_delete_renderbuffer(ctx, rb);
+      return NULL;
+   }
+
+   return irb;
+}
+
 void
 intel_renderbuffer_set_draw_offset(struct intel_renderbuffer *irb)
 {
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.h b/src/mesa/drivers/dri/intel/intel_fbo.h
index 9c48e9c1e8f..f135dead3ee 100644
--- a/src/mesa/drivers/dri/intel/intel_fbo.h
+++ b/src/mesa/drivers/dri/intel/intel_fbo.h
@@ -140,6 +140,10 @@ intel_create_wrapped_renderbuffer(struct gl_context * ctx,
 				  int width, int height,
 				  gl_format format);
 
+struct intel_renderbuffer *
+intel_create_fake_renderbuffer_wrapper(struct intel_context *intel,
+                                       struct gl_texture_image *image);
+
 extern void
 intel_fbo_init(struct intel_context *intel);
 
diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c
index c9cbcf480d5..5acdb42c3ac 100644
--- a/src/mesa/drivers/dri/intel/intel_tex_copy.c
+++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c
@@ -41,6 +41,9 @@
 #include "intel_fbo.h"
 #include "intel_tex.h"
 #include "intel_blit.h"
+#ifndef I915
+#include "brw_context.h"
+#endif
 
 #define FILE_DEBUG_FLAG DEBUG_TEXTURE
 
@@ -177,15 +180,28 @@ intelCopyTexSubImage(struct gl_context *ctx, GLuint dims,
                      GLint x, GLint y,
                      GLsizei width, GLsizei height)
 {
-   if (dims == 3 || !intel_copy_texsubimage(intel_context(ctx),
-                               intel_texture_image(texImage),
-                               xoffset, yoffset,
-                               intel_renderbuffer(rb), x, y, width, height)) {
-      fallback_debug("%s - fallback to swrast\n", __FUNCTION__);
-      _mesa_meta_CopyTexSubImage(ctx, dims, texImage,
-                                 xoffset, yoffset, zoffset,
-                                 rb, x, y, width, height);
+   struct intel_context *intel = intel_context(ctx);
+   if (dims != 3) {
+#ifndef I915
+      /* Try BLORP first.  It can handle almost everything. */
+      if (brw_blorp_copytexsubimage(intel, rb, texImage, x, y,
+                                    xoffset, yoffset, width, height))
+         return;
+#endif
+
+      /* Next, try the BLT engine. */
+      if (intel_copy_texsubimage(intel_context(ctx),
+                                 intel_texture_image(texImage),
+                                 xoffset, yoffset,
+                                 intel_renderbuffer(rb), x, y, width, height))
+         return;
    }
+
+   /* Finally, fall back to meta.  This will likely be slow. */
+   fallback_debug("%s - fallback to swrast\n", __FUNCTION__);
+   _mesa_meta_CopyTexSubImage(ctx, dims, texImage,
+                              xoffset, yoffset, zoffset,
+                              rb, x, y, width, height);
 }