This is the new register generation toolkit in use by nouveau.
As far as I know, this is the best register description toolkit in
existence, and you should use it too for your hardware :)
Thanks to Marcin Kościelnicki for inventing it and performing
invaluable reverse engineering work of nVidia chips.
The old swtnl code was broken by the new shader linkage support for
GLSL.
This is a rewrite of swtnl support, which should instead work properly,
be faster and more closer to the much more tested hardware pipeline.
Intuition != mathematics, so this time I actually worked out the right
formula for first order approximation of perspective interpolation.
Ironically, per quad divide actually makes things slower when compared
with per pixel divide -- probably because the divide hardware unit is
rarely used, whereas the multiply unit is typically already saturated
and the first order approximation imply more multiplications.
These are the non-trivial conversions that this function recognizes,
which was produced by u_format_compatible_test.c:
b8g8r8a8_unorm -> b8g8r8x8_unorm
a8r8g8b8_unorm -> x8r8g8b8_unorm
b5g5r5a1_unorm -> b5g5r5x1_unorm
b4g4r4a4_unorm -> b4g4r4x4_unorm
l8_unorm -> r8_unorm
i8_unorm -> l8_unorm
i8_unorm -> a8_unorm
i8_unorm -> r8_unorm
l16_unorm -> r16_unorm
z24_unorm_s8_uscaled -> z24x8_unorm
s8_uscaled_z24_unorm -> x8z24_unorm
r8g8b8a8_unorm -> r8g8b8x8_unorm
a8b8g8r8_srgb -> x8b8g8r8_srgb
b8g8r8a8_srgb -> b8g8r8x8_srgb
a8r8g8b8_srgb -> x8r8g8b8_srgb
a8b8g8r8_unorm -> x8b8g8r8_unorm
r10g10b10a2_uscaled -> r10g10b10x2_uscaled
r10sg10sb10sa2u_norm -> r10g10b10x2_snorm
State trackers and pipe drivers should be updated to take advantage of
this knowledge, e.g., in surface_copy.
Also, include the color buffer in the key. Not having it there
causes a tight knots in the logic to determine when it is OK or not
to discard previous color buffer contents.
This extra validation is very useful when working on the built-ins, but
in general overkill - the results should stay the same unless the
built-ins or ir_validate have changed.
Also, validating all the built-in functions in every test case makes
piglit run unacceptably slow.
We might want to copy them as color ones though.
Also works around crash in Unigine Heaven due to failing to allocate
a 64 MB temporary in GART for a CPU copy.
Unigine Heaven now works on nv40, albeit with very heavy glitches (with
the floating branch with render_hdr 0).
Actually, we may want to get rid of the x/y coordinates for linear
surfaces, and realign the origin from scratch if necessary, instead
of doing this "on-demand realignment".
This reverts commit 5ad74779ce.
Sorry, but I had to revert this.
Any commit which needlessly increases the number of temporaries is wrong.
More temporaries mean less shader performance because of reduced parallelism
and therefore less efficient latency hiding. In this case, there is possible
performance degradation of every shader which uses GL state variables.
I cannot accept this.
Those are:
- dead-code elimination
- constant folding
- peephole (mainly copy propagation)
- register allocation
There are some bugs which I need to track down.
Also fix up the descriptions of all the debug options.
First list compiler passes in an array, then run the new function rc_run_compiler.
Every backend may need a different set of passes.
This cleans up the mess in r3xx_compile_vertex_program.
Luca Barbieri
José Fonseca
Kenneth Graunke
Marek Olšák
Tom Stellard