Everything is already in place; we simply have to take the scalar code
generation path. This gives us SIMD8 VS programs, instead of SIMD4x2.
v2: Rebase on the patch that drops brw->gen >= 8.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
I need to use this in brw_vec4.cpp, so it can't be static anymore.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Use prog_to_nir where we would normally call glsl_to_nir, handle program
parameter lists, and skip a few things that don't exist.
Using NIR generates much better shader code than Mesa IR, since we get
real optimizations, as opposed to prog_optimize:
total instructions in shared programs: 314007 -> 279892 (-10.86%)
instructions in affected programs: 285173 -> 251058 (-11.96%)
helped: 2001
HURT: 67
GAINED: 4
LOST: 7
v2: Change early return in nir_setup_uniforms to if/else (Jordan).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
prog->nir will generate fsub opcodes, but i965 doesn't implement them.
We may as well lower them at the NIR level, since it's trivial to do.
Suggested by Connor Abbott.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Shamelessly ripped off from Eric Anholt's tgsi_to_nir pass.
This is not built on SCons, like the rest of NIR.
v2:
- Delete redundant c->s, c->impl, and c->cf_node_list pointers (Ken)
- Use nir_builder directly instead of ptn_compile in more places (Ken)
- Drop 'struct' keyword in front of nir_builder (ken)
- Add a file level Doxygen comment (Ken)
- Use scalar constants instead of splatting (Eric)
- Use nir_builder helpers for constants, moves, and swizzles (Connor)
v3: Minor indentation improvements.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
These will be useful for prog->nir and tgsi->nir.
v2: Don't forget to mark nir_swizzle as inline (Eric).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
The PMA depth stall must be enabled (optimization turned off) under certain
circumstances on gen8. This was supposedly fixed for Gen9, which means we do not
need to check, or toggle the state. The hardware is supposed to enable the
hardware optimization by default, unlike BDW, so we also don't need to set it at
init. For whatever reason this improves stability on ETQW with the bug mentioned
below.
References: https://bugs.freedesktop.org/show_bug.cgi?id=89039 (doesn't fix)
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: Anuj Phogat <anuj.phogat@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Recomendation [sic] is to set this field to 1 always. Programming it to default
value of 0, may have -ve impact on performance for MSAA WLs.
Another don't suck bit which needs to get set.
The patch wasn't as well tested as I would have liked, primarily I don't have
perf numbers for it, but it's getting to a point where it is in danger of being
lost.
v2: v1 was a mix of two patches. Since 0x7004 is masked, we only need to set it
once at initialization and make sure the pma workaround doesn't set the mask bit
(which it doesn't).
Move LRI to init gpu state (Ken)
Add a comment.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
These functions looked quite complicated, even though what they actually did
was trivial (ever since we dropped swizzled rendering). Also drop lookup of
format block per bytes done for each block, and do it once per scene instead.
This improves everybody's favorite "benchmark" by 3% or so, though
lp_rast_shade_quads_all() which calls this shows up still quite high for a
function which does little more than call the jit function.
(This would most likely be much better handled by the jit function itself,
the strides are passed through anyway already, though for being able to
handle layers it would definitely add some complexity.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
When using the texel fetch functions rather than ordinary texturing,
the arguments are all int vecs instead of float vecs, not to mention
the actual function would look completely different. Hence this must
be included in the texture function name (which serves as the key)
otherwise things crash badly when a shader accesses the same texture
and sampler unit with both txf/ld and ordinary texturing instructions
with otherwise matching keys.
There are issues with inlining everything, most notably llvm will use much
more memory (and be slower) when compiling. Ideally we'd probably use
functions for shader functions too but texture sampling usually is responsible
for quite some IR (it can easily reach 80% of total IR instructions) so this
seems like a good start.
This still generates a different function for all different combinations just
like before, however it is possible llvm is missing some optimization
opportunities - it is believed though such opportunities should be somewhat
rare, but at least for now it can still be switched off (at compile time only).
It should probably make compiled code also smaller because the same function
should be used for different variants in the same module (so for the
opaque/partial or linear/elts variants).
No piglit change (though it does indeed speed up unrealistic tests like
fp-indirections2 by a factor of 30 or so).
Has a small negative performance impact in openarena - I suspect this could
be fixed by running some IPO passes (despite the private linkage, llvm right
now does NO optimization at all wrt anything going past the call, even if
there's just one caller - so things like values stored before the call and then
always written by the function etc. will not be optimized away, nor will dead
arguments (which we mostly shouldn't have) be eliminated, always constant
arguments promoted etc.).
v2: use proper return values instead of pointer function arguments.
llvm supports aggregate return values, which do wonders here eliminating
unnecessary stack variables - everything in fact will be returned in registers
even without any IPO optimizations. It makes the code simpler too.
With this I could not measure a peformance impact in openarena any longer
(though since there's still no constant value propagation etc. into the tex
functions this does not mean it couldn't have a negative impact elsewhere).
v3: fix some minor issues suggested by Jose, and do disassembly (and the
profiling) without hacks.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The callbacks used for getting the dynamic texture/sampler state were using
the jit_context from the generated jit function. This works just fine, however
that way it's impossible to generate separate functions for texture sampling,
as will be done in the next commit. Hence, pass this pointer through all
interfaces so it can be passed to a separate function (technically, it would
probably be possible to extract this pointer from the current function instead,
but this feels hacky and would probably require some more hacks if we'd use
real functions instead of inlining all shader functions at some point).
There should be no difference in the generated code for now.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The data in memory is in big endian format and needs to be converted
into CPU byte order. So the patch actually reversed what needs to be done.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The CUBE_ARRAY case uses r[4]. Make sure that the stack variable is
there.
Noticed by Coverity.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Does not appear to be used in tree. Coverity spotted some errors in the
bitmask stuff, but the whole thing appears to be unused.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Brian Paul <brianp@vmware.com>
Our fragment program backend implements support for TXP directly, and
there's no NIR lowering pass to remove the projection. When we switch
fragment program support over to NIR, we need to support it somehow.
It's easy enough to support directly.
v2: Split out offset/tex_offset rename (requested by Jordan).
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
fs_visitor::nir_emit_texture() created an fs_reg variable called offset,
which shadowed the offset() helper function in brw_ir_fs.h.
Rename the variable to tex_offset so we can still call offset().
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
v2: Don't be lazy. Constify the as_foo functions and use those instead
of ugly casts. Suggested by Curro.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Now that they're all implemented using macros, this is trivial.
v2: Remove redundant parenthesis. Suggested by Curro.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
The downcast functions for non-leaf classes were previously implemented
"by hand." Now they are implemented using macros based on the is_foo
functions added in the previous patch.
v2: Remove redundant parenthesis. Suggested by Curro (on the next
patch).
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
These functions deteremine when an IR node is one of the non-leaf
classes.
v2: Adjust indentation to line up. Suggested by Matt.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
These are due how we implemented the atomic tests, not the atomic
implementation itself. It's also difficult to refactor the code to
avoid the warnings due to the use of macros -- the code would be quite
hairy.
Reviewed-by: Brian Paul <brianp@vmware.com>
MSVC's implementation of signbit(x) uses sizeof(x) with expressions to
dispatch to an internal function based on the argument's type (float,
double, etc), but that raises a flag with MSVC's own static analyzer,
and because this is an inline function in a header it causes substantial
warning spam.
Reviewed-by: Brian Paul <brianp@vmware.com>
There doesn't seem much interest on osmesa on Windows, particularly classic osmesa.
If there is indeed interest in osmesa on Windows, we should instead
integrate src/gallium/targets/osmesa into SCons.
Reviewed-by: Brian Paul <brianp@vmware.com>
v2: Review from Laura Ekstrand
- get rid of a change that should not have happened in this patch
- improve the error messages
- fix alignments
- fix a capitalization in a function name in an error message
v3: Review from Laura Ekstrand
- move the test for the validity of the renderbuffer to less generic
functions
- get rid of some changes that accidentally landed in the wrong commit
- revert some alignment fixes
v3: Review from Laura Ekstrand
- check that the lookup returns a valid renderbuffer
- cosmetic changes to some error messages
Reviewed-by: Laura Ekstrand <laura@jlekstrand.net>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
v2:
- improve an error message
v3:
- move a test to less generic functions
- fix an alignment
v4:
- take the caller as a parameter instead of bool dsa
- check that the lookup returns a valid renderbuffer
Reviewed-by: Laura Ekstrand <laura@jlekstrand.net>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
Because of the current way the code is architectured, there is no
functional difference between the DSA and the non-DSA path.
Reviewed-by: Laura Ekstrand <laura@jlekstrand.net>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
These entry points will be fleshed out when the GL_ARB_query_buffer_object
extension gets implemented. In the meantime, return GL_INVALID_OPERATION as
suggested by Ian.
Reviewed-by: Laura Ekstrand <laura@jlekstrand.net>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
v2: Review from Laura Ekstrand
- use the transform feedback object lookup wrapper
v3:
- use the new name of _mesa_lookup_transform_feedback_object_err
v4: Review from Laura Ekstrand
- fix some alignement problems
Reviewed-by: Laura Ekstrand <laura@jlekstrand.net>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
v2: Review from Laura Ekstrand
- use the transform feedback object lookup wrapper
v3:
- use the new name of _mesa_lookup_transform_feedback_object_err
Reviewed-by: Laura Ekstrand <laura@jlekstrand.net>
Signed-off-by: Martin Peres <martin.peres@linux.intel.com>
Kenneth Graunke