Age | Commit message (Collapse) | Author |
|
|
|
|
|
is dificult to quantify but the improvemnet reaches about of 0.7% in some cases in a Pentium III. In older Pentiums the improvements could be quite substancial due less capable integer pipelines.
|
|
Slight optimization of _mesa_mmx_blend_add.
|
|
|
|
Factorization of more common MMX code.
|
|
as scaling and lerping.
Restructured the MMX blending function to use a template, being only necessary to specify the main loop, which is also used for making the runin and runout sections.
Optimization of the MMX function after remembering that the multiplication was commutative (how can somebody forget this..) resulting in less register usage. Now there is no need for generate or read from memory any constant inside the loop.
Assemblers other than the GNU assembler can choke on the output of the C preprocessor since it was necessary to add line separators ';' to the defined macros.
|
|
complexity and facilitate reusability.
|
|
made also in the runin and runout section of the code.
|
|
|
|
|
|
|
|
|
|
matypes.h, which includes assyntax.h and is generated from the core Mesa
header files.
|
|
|
|
general cleanups - more to come.
Added P6 architecture timing to debug_xform routines. Still need to add
test_all_vertex_functions test for the v16 asm. Dynamic reconfiguration
of counter overhead for more accurate benchmarking.
|
|
removed spaces from all uses of LLBL macro, for FreeBSD
|
|
|