Salvia Optimization

[GoogleCodeExporter]

Description:
  size of shader registers (vsi vso psi pso) is not fixed.
  That may improve performance.

Original issue reported on code.google.com by wuye9036 on 24 Apr 2014 at 11:49

[GoogleCodeExporter]

This issue was updated by revision d6bddce56b2e.

Added vls_* code.

Original comment by wuye9036 on 24 Apr 2014 at 11:52

[GoogleCodeExporter]

This issue was updated by revision 2285a941ed77.

Rename shaderregs and shaderregs_op to shader_regs and shader_regs_op.
Remove register operations to shader_regs_op.
Remove shader_object.cpp which is unused.
Add code to make register_file.
Rename preserved_pool to reserved_pool.

Original comment by wuye9036 on 27 Apr 2014 at 3:22

[GoogleCodeExporter]

This issue was updated by revision 0749aab73fa5.

Add pointer_calc to eflib.

Original comment by wuye9036 on 27 Apr 2014 at 3:22

[GoogleCodeExporter]

This issue was updated by revision 3e0b08187ba0.

Fixed stride init bug in pool.

Original comment by wuye9036 on 27 Apr 2014 at 9:22

[GoogleCodeExporter]

This issue was updated by revision 51c0dcf2eccb.

Extract function compute front from clip.
Fixed clipping quick test errors.

Original comment by wuye9036 on 27 Apr 2014 at 9:22

[GoogleCodeExporter]

This issue was updated by revision d532f783acd2.

Add parallel_sort support to unique vertex sorting.
Disable _SECURE macros in CMake but not in config.h
Add experimental code for optimize interpolation code.

Original comment by wuye9036 on 30 Apr 2014 at 12:55

[GoogleCodeExporter]

This issue was updated by revision 5683408033b2.

Update peformance documents.

Original comment by wuye9036 on 30 Apr 2014 at 5:22

[GoogleCodeExporter]

This issue was updated by revision e55e48e43265.

interpolate function works on quad. It reduced mul and add operations.

Original comment by wuye9036 on 30 Apr 2014 at 5:22

[GoogleCodeExporter]

This issue was updated by revision de2ba3d68b56.

cpp PS works with quad pixels.
Now cpp PS ddx/ddy is computed by quad attributes substraction.
Remove unused sample and tex* functions.

Original comment by wuye9036 on 1 May 2014 at 1:03

[GoogleCodeExporter]

This issue was updated by revision a5c8062f298a.

Remove unused shader register ops.

Original comment by wuye9036 on 1 May 2014 at 1:03

[GoogleCodeExporter]

This issue was updated by revision d2b4568ac22a.

Remove front_face attribute from vs_output and ps_output.
front face is set before PS execution.

Original comment by wuye9036 on 1 May 2014 at 1:03

[GoogleCodeExporter]

This issue was updated by revision 9f4132cadf79.

Add texture.get_int_size to reduce virtual function call.

Original comment by wuye9036 on 1 May 2014 at 1:03

[GoogleCodeExporter]

This issue was updated by revision d4069cb20852.

Refactor texture.
CANNOT BE COMPILED!

Original comment by wuye9036 on 1 May 2014 at 1:03

[GoogleCodeExporter]

This issue was updated by revision e1de2a75f01d.

Update performance document.

Original comment by wuye9036 on 1 May 2014 at 7:31

[GoogleCodeExporter]

This issue was updated by revision c57caab9a779.

Refactor surface and sampler interfaces done.
Update code line stats.

Original comment by wuye9036 on 1 May 2014 at 4:17

[GoogleCodeExporter]

This issue was updated by revision c9f4c4c84d64.

update performance document.

Original comment by wuye9036 on 1 May 2014 at 4:17

[GoogleCodeExporter]

This issue was updated by revision 525df7da4539.

Optimize triangle derivative computation.
Now viewport and projection tranformation run as parallel.

Original comment by wuye9036 on 1 May 2014 at 11:43

[GoogleCodeExporter]

This issue was updated by revision f784c33356d5.

Remove unused functions in shader_op.
Reduce MAX_VS_OUTPUT_ATTRS from 8 to 7.

Original comment by wuye9036 on 2 May 2014 at 1:41

[GoogleCodeExporter]

This issue was updated by revision 4f059cff7edc.

transformed_verts_ in default_vertex_cache now reused between draw calls.

Original comment by wuye9036 on 2 May 2014 at 4:18

[GoogleCodeExporter]

This issue was updated by revision 22f25e15ab3f.

Update performance document.

Original comment by wuye9036 on 2 May 2014 at 4:18

[GoogleCodeExporter]

This issue was updated by revision c07ce4b2a332.

Fixed a conditional compiling error.

Original comment by wuye9036 on 2 May 2014 at 4:19

[GoogleCodeExporter]

This issue was updated by revision 41083fb6a950.

Added max sample count and sample count mask to renderer_capacity.
Add render_sample_quad to frame buffer.

Original comment by wuye9036 on 7 May 2014 at 4:15

[GoogleCodeExporter]

This issue was updated by revision 7ca99cb789bb.

Now early_z_test returns mask but not true/false.
Add early_z_test_quad.

Original comment by wuye9036 on 7 May 2014 at 4:15

[GoogleCodeExporter]

This issue was updated by revision e447f68f3e6d.

pixel shader now use 2x2 quad to instead 4x4.
Added step_2d_unproj_pos_quad.

Original comment by wuye9036 on 7 May 2014 at 8:17

[GoogleCodeExporter]

This issue was updated by revision f76bba75ddbe.

Now rendering is based on 2x2 quads.

Original comment by wuye9036 on 7 May 2014 at 8:17

[GoogleCodeExporter]

This issue was updated by revision 771d0a142bfd.

Re-enable early z tests.

Original comment by wuye9036 on 7 May 2014 at 8:17

[GoogleCodeExporter]

This issue was updated by revision b08b28f2195a.

Move vs_output of quad to draw quad functions.

Original comment by wuye9036 on 8 May 2014 at 3:14

[GoogleCodeExporter]

This issue was updated by revision 59a680c6842d.

Reuse geom_setup_engine in rasterizer
Reuse clipped verts buffer.
Remove depth and coverage field from pso.
Improve clipped vertex copy if clipping is not occurred.

Original comment by wuye9036 on 17 May 2014 at 10:19

[GoogleCodeExporter]

This issue was updated by revision efefd7c92302.

Fixed bug in cpp_pixel_shader::ddy().

Original comment by wuye9036 on 17 May 2014 at 10:19

[GoogleCodeExporter]

This issue was updated by revision 11be6fcf43f5.

Add pipeline_profiles to async object.

Original comment by wuye9036 on 19 May 2014 at 4:20

[GoogleCodeExporter]

This issue was updated by revision 895ca0a19ef4.

Finished profiling in rendering pipeline.
PipelineStat used profiling.

Original comment by wuye9036 on 19 May 2014 at 10:11

[GoogleCodeExporter]

This issue was updated by revision e2299785cf9f.

Now generate index could return min/max of indexes.

Original comment by wuye9036 on 20 May 2014 at 3:10

[GoogleCodeExporter]

This issue was updated by revision b9158945e790.

Separate clipping and compact clip in profiles.
Finished USE_INDEX_RANGE. but it is not better than old version, just make it 
disabled in default.
Add clip_triangle_to_poly_simple but not used.

Original comment by wuye9036 on 20 May 2014 at 7:33

[GoogleCodeExporter]

This issue was updated by revision f930d4cf1eae.

Improved reserved_pool.
Eliminate unnecessary malloc/free in geom_setup_engine to optimization.
Rename transform_vertices to prepare_vertices.
Add fetch3 to vertex cache which can fetch 3 verts once time.
Add compact clip profile display to PipelineStat.
Remove some warnings.

Original comment by wuye9036 on 21 May 2014 at 7:45

[GoogleCodeExporter]

This issue was updated by revision f880dd44cb03.

Fixed a false assertion in reserved_pool.
Add vs_output_attr_count to host.
Remove code referenced old vertex shader unit.
vertex cache preparation is moved into raster draw.
RUN-TIME ERROR!

Original comment by wuye9036 on 22 May 2014 at 3:47

[GoogleCodeExporter]

This issue was updated by revision e864158db3ee.

Fixed bug of interpolation method initialization before draw.

Original comment by wuye9036 on 22 May 2014 at 5:58

[GoogleCodeExporter]

This issue was updated by revision dea6302b2d70.

Refactor vertex_cache, now it supports multi-threaded vertex fetch.
Extract default implements to vertex_cache_impl.

Original comment by wuye9036 on 22 May 2014 at 8:04

[GoogleCodeExporter]

Original comment by wuye9036 on 22 May 2014 at 9:05

• Changed title: _Salvia Optimization _

[GoogleCodeExporter]

This issue was updated by revision 2a538328848c.

For early-z case, depth read-test-write are executed together..

Original comment by wuye9036 on 30 May 2014 at 10:52

[GoogleCodeExporter]

This issue was updated by revision 3d949b15b08f.

Rename default_vertex_cache to precomputed_vertex_cache
Add shared_vertex_cache.

Original comment by wuye9036 on 30 May 2014 at 10:52

[GoogleCodeExporter]

This issue was updated by revision 37c244d9a71a.

Add simple statistics into shared_vertex_cache for: spin lock conflict count; 
L2 hit/miss ratio.
Improve spin lock in vertex cache.

Original comment by wuye9036 on 30 May 2014 at 10:52

[GoogleCodeExporter]

This issue was updated by revision 0dee9f310692.

Merged shared index and shared vso.
It improved performance obviously, maybe due to less false sharing.

Original comment by wuye9036 on 30 May 2014 at 11:21

[GoogleCodeExporter]

Original comment by wuye9036 on 4 Jun 2014 at 6:10

• Changed state: Fixed