[Mesa-dev] [PATCH 00/59] Enable GL_ARB_gpu_shader_int64

[Ian Romanick]

Whew!  Here it is!  There are four main groups to this patch series.
All of this is available at:

    https://cgit.freedesktop.org/~idr/mesa/log/?h=arb_gpu_shader_int64

There are some gaps in our current piglit tests.  For example, I
discovered that there are no execution tests for bitwise not... when I
discovered that it didn't work. :)  This series does pass all the tests
to which I have access, so it feels safe, especially this early in the
17.0 release process, to enable the extension now.  I expect the testing
situation will impove in the next few months.

Here's the rough breakdown of the series:

Patches 1 through 7: Add support to the standalone compiler to generate
C++ that uses ir_builder to generate the compiled shader.  I sent a
version of this out previously, but there have been some functional
changes.  This is used later in the series to generate some comples
built-in functions.

Patches 8 through 32: Mostly Dave's original code that enables the core
GL_ARB_gpu_shader_int64 functionality.  I have reviewed or modified all
of Dave's patches, but additional eyes are quite welcome.

Patches 33 through 43: Enable additional lowering functions that i965
will need.  I suspect that most drivers will need to use the
multiplication and division functions.  Radeon does not need these
because the lowering is done in LLVM.

These functions are also made available, with #defines, when
GL_MESA_shader_integer_functions is enabled.  This will allow testing
the lowering functions even on drivers that don't use them.  This will
help prevent cases where changes to the core compiler or optimization
passes cause problems for drivers that aren't under test.

Patches 44 through 50: Enable 64-bit integer support in NIR.

Patches 50 through 59: Enable remaining bits ofr 64-bit integer support
in the i965 compiler backend.  "i965: Really don't emit Q or UQ moves on
Gen < 8" and "i965: Split SIMD16 CMP of Q and UQ instructions" went
through numerous iterations.  If there are suggestions for handling
these better, I would like to hear them.