summaryrefslogtreecommitdiff
AgeCommit message (Collapse)Author
2007-12-14[intel] Remove the relocation buffer lists and just cache one per buffer.Eric Anholt
Each buffer object now has a relocation buffer pointer, which contains the relocations for the buffer if there are any. At the point where we have to create a new type of relocation entry, we can change the code over to allowing multiple relocation lists, but trying to anticipate what that'll look like now just increases complexity. This is a 30% performance improvement on 965.
2007-12-14[965] Replace the state cache suballocator with direct dri_bufmgr use.Eric Anholt
The user-space suballocator that was used avoided relocation computations by using the general and surface state base registers and allocating those types of buffers out of pools built on top of single buffer objects. It also avoided calls into the buffer manager for these small state allocations, since only one buffer object was being used. However, the buffer allocation cost appears to be low, and with relocation caching, computing relocations for buffers is essentially free. Additionally, implementing the suballocator required a don't-fence-subdata flag to disable waiting on buffer maps so that writing new data didn't block on rendering using old data, and careful handling when mapping to update old data (which we need to do for unavoidable relocations with FBOs). More importantly, when the suballocator filled, it had no replacement algorithm and just threw out all of the contents and forced them to be recomputed, which is a significant cost. This is the first step, which just changes the buffer type, but doesn't yet improve the hash table to not result in full recompute on overflow. Because the buffers are all allocated out of the general buffer allocator, we can no longer use the general/surface state bases to avoid relocations, and they are set to 0 instead.
2007-12-13[intel] Remove broken mutex protection from dri_bufmgrs.Eric Anholt
Now that the dri_bufmgr is stored in the context rather than the screen, all access to one is single-threaded anyway.
2007-12-13[intel] Enable INTEL_DEBUG=bufmgr output in TTM mode as well as classic.Eric Anholt
2007-12-13[intel] assert that buffers are not mapped at last unreference.Eric Anholt
bufmgr_fake doesn't care about it, but with ttm we would end up with the buffer remaining referenced until application exit.
2007-12-12[intel] Assert against conflicting relocation emits in bufmgr_fake.cEric Anholt
2007-12-12[965] Bug #13600: Fix assertion failure with SRGB textures.Eric Anholt
I broke this with cherry-pick resolving on 93c98a466947570e0589b662df49095b2f4bc43c.
2007-12-12[intel] Move bufmgr back to context instead of screen, fixing glthreads.Eric Anholt
Putting the bufmgr in the screen is not thread-safe since the emit_reloc changes. It also led to a significant performance hit from pthread usage for the attempted thread-safety (up to 12% of a cpu spent on refcounting protection in single-threaded 965). The motivation had been to allow multi-context bufmgr sharing in classic mode, but it wasn't worth the cost.
2007-12-11Block in kernel waiting for fenceKeith Packard
2007-12-11Use previous buffer offsets to compute proposed relocationsKeith Packard
This takes advantage of the DRM_BO_HINT_PRESUMED_OFFSET change and allows the kernel to avoid mapping and re-writing buffers when relocations occur.
2007-12-11add simple test for two-sided stencil functionalityRoland Scheidegger
2007-12-11make sure state token values are fully initializedRoland Scheidegger
2007-12-10[965] Hook up DEBUG_BUFMGR output for bufmgr_fake.Eric Anholt
2007-12-07[965] Convert the driver to dri_bufmgr interface and enable TTM.Eric Anholt
This is currently believed to work but be a significant performance loss. Performance recovery should be soon to follow. The dri_bo_fake_disable_backing_store() call was added to allow backing store disable like bufmgr_fake.c did, which is a significant performance win (though it's missing the no-fence-subdata part). This commit is a squash merge of the 965-ttm branch, which had some history I wanted to avoid pulling due to noisiness and brokenness at many points for git-bisecting.
2007-12-07[965] Remove dead code in upload_wm_surfaces.Eric Anholt
2007-12-07[965] Move brw_surface_state stack allocation into the function using it.Eric Anholt
2007-12-07Make osdemos linking like other programsDan Nicholson
Most of the programs list their dependencies on the Mesa libraries in their Makefiles. This works with the default configuration where APP_LIB_DEPS only lists external libraries. This changes the linux-osmesa configs and the osdemos Makefile to follow that convention. Some cleanup of the Makefile is also added to refer to the GL libraries through the existing variables rather than hardcoding their names.
2007-12-07configs: Fix linking with static libGL and --as-neededDan Nicholson
Linking of the programs breaks when using a static libGL and the GNU ld option --as-needed. This is because libXext is needed for the XShm functions.
2007-12-07i915: fix the error in the previos commit.Xiang, Haihao
2007-12-07i915: Check the program size when uploading a program. fix bug 13494Xiang, Haihao
2007-12-06Minor followup fixes for the previous commit.Michel Dänzer
2007-12-06Always call dlopen in DriverOpen.George Nassas
This increases the reference count for the driver binary, preventing it from getting unloaded prematurely in driDestroyDisplay. See https://bugs.freedesktop.org/show_bug.cgi?id=13541 .
2007-12-05Revert "[965] Add missing flagging of new stage programs for updating stage ↵Eric Anholt
state." I had forgotten part of brw_state_cache.c that made this fix not relevant for master (last_addr comparison and flagging based on cache id). This reverts commit a4642f3d18bdaebaba31e5dee72fe5de9d890ffb.
2007-12-05[965] Add missing flagging of new stage programs for updating stage state.Eric Anholt
Otherwise, choosing a new program wouldn't necessarily update the state, and and an old program could be executed, leading to various sorts of pretty pictures or hangs.
2007-12-05Fix endianness bug in _mesa_texstore_argb8888()Brian
On big-endian, storing in _mesa_texformat_argb8888 format produced wrong res Also, clean-up nearby code to match. picked from gallium-0.1 branch
2007-12-05Don't Swap buffer if a DRIDrawable is entirely obscuredXiang, Haihao
by another window.
2007-12-04Undo changes made to the toyball shaders in commit ↵Brian
d19d0596daf004b56d80f78fa1a329b43c2ebf94 This demo produces the expected results again.
2007-12-04fix span->facing computation and gl_FrontFacing initializationBrian
2007-12-04Fix gl_FrontFacing compilation problemBrian
2007-12-04added culling/wireframe optionsBrian
2007-12-04Use -Bsymbolic for linking all shared objects.Michel Dänzer
Fixes https://bugs.freedesktop.org/show_bug.cgi?id=10132 . Also remove comment about SONAME, as SONAME only applies to shared libraries.
2007-12-03[965] Change constant buffer from state structs to plain batch emission.Eric Anholt
Reduces diff to branch which has a relocation in this state emit.
2007-12-03fix DD_TRI_LIGHT_TWOSIDE bug (#13368)Brian
2007-12-03Fix the library name in glw.pcDan Nicholson
Fix a copy and paste error s/GLU/GLw/ in glw.pc.
2007-12-03glut doesn't need -lXtBrian
2007-12-03added missing quote charBrian
2007-12-03i915: Fix up state changes for i8xx.Michel Dänzer
2007-11-30[intel] Move batch bo_unmap from TTM code to shared, and add more asserts.Eric Anholt
2007-11-30[intel] Add failure path printfs to relocation code and some comments.Eric Anholt
2007-11-30[intel] Simplify TTM relocation code by passing around bufmgr struct.Eric Anholt
2007-11-30[intel] Fix the type and naming of the flags/mask args to TTM functions.Eric Anholt
The uint64_t flags (as defined by drm.h) were being used as unsigned ints in many places.
2007-11-30[intel] intel_bufmgr_ttm style sanityEric Anholt
2007-11-30fix-build: remove ctx->_Facing assignmentBrian
2007-11-30fix broken two-sided stencilBrian
2007-11-30better front-plane clip testBrian
2007-11-30Updates of some OpenVMS makefiles.joukj
2007-11-30Merge branch 'master' of git+ssh://joukj@git.freedesktop.org/git/mesa/mesajoukj
2007-11-30i965: if source depth to render target is set,Xiang, Haihao
it should be handled in fb_write.
2007-11-30i965: use uncompressed instruction to ensure onlyXiang, Haihao
Pixel Mask Copy is modified as the pixel shader thread turns off pixels based on kill instructions.
2007-11-29[i915] Make INTEL_DEBUG=bufmgr actually do things for bufmgr_fake.Eric Anholt