summaryrefslogtreecommitdiffstats
Commit message (Collapse)AuthorAgeFilesLines
* Implement pow/rcp and sub opcodesZack Rusin2007-10-243-5/+91
|
* Add the "add" opcode and assert one vp if an opcode isn't supportedZack Rusin2007-10-241-1/+9
|
* Refactor the tgsi->llvm storage translatorZack Rusin2007-10-244-42/+71
|
* Implement basic opcode translation and storage translation.Zack Rusin2007-10-245-27/+350
|
* Draw first triangle. Start on the llvm builder.Zack Rusin2007-10-243-24/+237
|
* Cleanup some of the testing code. Implement first pass at actuallyZack Rusin2007-10-245-42/+208
| | | | running shaders in llvm.
* Execution engine is a singleton, for now keep it in the pipe.Zack Rusin2007-10-246-19/+64
|
* Generate the base shader.Zack Rusin2007-10-244-47/+194
|
* Implement the conversion and do the initial execution pass.Zack Rusin2007-10-2410-11/+244
|
* Stub out some conversion.Zack Rusin2007-10-241-0/+291
|
* Initial stab at LLVM integration.Zack Rusin2007-10-248-5/+225
|
* Fix vertex cache bug that allows multiple vertices to fall into the same slot.Brian2007-10-231-0/+9
| | | | | | | | | | | | Need to set the slot's bit in draw->vcache.referenced even when there was a cache hit since flushing the primitive buffer will have cleared the bitfield but not the cache's vertex indexes. Fixes a bug found when drawing long triangle fans but could be hit by other prim types as well. An alternate fix would be to call draw_vertex_cache_invalidate() from draw_vertex_cache_unreference().
* added commentBrian2007-10-231-1/+1
|
* In get_vertex(), slot was computed using & 31. Replace with % VCACHE_SIZE.Brian2007-10-231-1/+3
| | | | Also, assert that index is not too large before indexing array.
* adjust coords in wide_line() to be conformantBrian2007-10-231-8/+47
|
* properly init dst reg's CondMask/Swizzle fieldsBrian2007-10-232-5/+5
|
* bump up MAX_INSTRUCTIONS and add an assertion to catch emitting too many ↵Brian2007-10-231-3/+6
| | | | instructions
* Don't pad renderbuffers to multiple of two pixels anymore.Brian2007-10-231-6/+1
| | | | | This was only needed to avoid out-of-bounds memory accesses with the 2x2 quad_read/write() functions which no longer exist.
* don't apply ColorMask to main gcBrian2007-10-221-7/+6
|
* fix masking bug, memory leakBrian2007-10-221-29/+28
|
* don't use GL types for get/put_tile() paramsBrian2007-10-221-4/+4
|
* new flag to control psize (from vertex shader or fixed size)Brian2007-10-223-1/+26
|
* add support for sprite texcoord modesBrian2007-10-224-7/+40
|
* implement point sprite modeBrian2007-10-222-35/+48
|
* add point_sprite flag to rasterizer stateBrian2007-10-222-0/+2
|
* remove unused varBrian2007-10-221-1/+1
|
* tweak point corners to pass conform testBrian2007-10-221-4/+8
|
* plug the wide prims code into the pipelineBrian2007-10-224-1/+11
|
* update to working conditionBrian2007-10-221-111/+91
|
* Finish unifying the surface and texture tile caches.Brian2007-10-228-118/+72
|
* pull clip/ module wide and stippled lines/points codekeithw2007-10-222-0/+541
|
* Start implementing cache routines for textures.Brian2007-10-212-1/+71
| | | | First step to consolidating surface/texture caching...
* rename some varsBrian2007-10-211-25/+25
|
* silence warningBrian2007-10-212-1/+2
|
* Remove obsolete read/write_quad() functionsBrian2007-10-203-591/+54
|
* init sp->sbuf_cache to avoid possible segfaultBrian2007-10-201-0/+1
|
* Call softpipe_unmap_surfaces() in softpipe_flush().Brian2007-10-201-0/+8
| | | | | | | This fixes a DRM BO failure upon swapbuffers caused by the color buffer still being mapped. This is a bit heavy handed since we don't always need to unmap buffers when flushing. Need to pass a flag to flush() or design a new function.
* In region_unmap(), check if region is mapped before decrementing refcount.Brian2007-10-202-8/+14
|
* flush the pipe before accum opsBrian2007-10-201-0/+3
|
* unmap regions when reallocating renderbuffer storageBrian2007-10-201-0/+4
|
* renderbuffer tweaks in update_framebuffer_state()Brian2007-10-201-4/+5
|
* use combined depth/stencil buffer when possibleBrian2007-10-201-6/+16
|
* Convert Z/stencil ops to use cached tiles like colors.Brian2007-10-2016-126/+395
| | | | | Also, quite a bit of re-org of the tile caches and surface mapping/unmapping. Leave surfaces mapped between primitives now.
* added case for TGSI_OPCODE_ENDBrian2007-10-201-0/+4
|
* get/put_tile_raw() funcs for 16/32bpp surfacesBrian2007-10-191-0/+115
|
* added get/put_tile_raw() methodsBrian2007-10-191-4/+17
|
* don't alloc region in xmesa_surface_alloc(), fixes a mem leakBrian2007-10-191-3/+1
|
* disable debug printfBrian2007-10-191-2/+1
|
* call Driver.Flush() in _mesa_notifySwapBuffers()Brian2007-10-191-2/+4
|
* Initial implementation of surface tile caching.Brian2007-10-199-72/+129
| | | | | | Instead of using read/write_quad() functions, do framebuffer accesses via get/put_tile(). A cache of tiles is used to avoid frequent get/put() calls. Only implemented for color buffers right now.