nir: Add a "compact array" flag and IO lowering code.
authorKenneth Graunke <kenneth@whitecape.org>
Tue, 4 Oct 2016 03:32:22 +0000 (20:32 -0700)
committerKenneth Graunke <kenneth@whitecape.org>
Tue, 22 Nov 2016 08:29:23 +0000 (00:29 -0800)
commit663b2e9a92f152c314f611526b8a16ff4c06249f
tree5be196955fca631312a0561fcad7c116615973f4
parentf395e3445df98b0398865a1897bca003f9add6f4
nir: Add a "compact array" flag and IO lowering code.

Certain built-in arrays, such as gl_ClipDistance[], gl_CullDistance[],
gl_TessLevelInner[], and gl_TessLevelOuter[] are specified as scalar
arrays.  Normal scalar arrays are sparse - each array element usually
occupies a whole vec4 slot.  However, most hardware assumes these
built-in arrays are tightly packed.

The new var->data.compact flag indicates that a scalar array should
be tightly packed, so a float[4] array would take up a single vec4
slot, and a float[8] array would take up two slots.

They are still arrays, not vec4s, however.  nir_lower_io will generate
intrinsics using ARB_enhanced_layouts style component qualifiers.

v2: Add nir_validate code to enforce type restrictions.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
src/compiler/glsl/glsl_to_nir.cpp
src/compiler/nir/nir.h
src/compiler/nir/nir_gather_info.c
src/compiler/nir/nir_lower_indirect_derefs.c
src/compiler/nir/nir_lower_io.c
src/compiler/nir/nir_print.c
src/compiler/nir/nir_validate.c