| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
| |
Reduces vertex shader instruction counts in DOTA2 by 6.42%, L4D2 by
4.61%, and CS:GO by 5.71%.
total instructions in shared programs: 1500153 -> 1498191 (-0.13%)
instructions in affected programs: 59919 -> 57957 (-3.27%)
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
| |
Now that loop_controls no longer creates normatively bound loops,
there is no need for ir_loop::normative_bound or the
lower_bounded_loops pass.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, all of the back-ends (ir_to_mesa, st_glsl_to_tgsi, and the
i965 fs and vec4 visitors) had nearly identical logic for handling
bounded loops. This replaces the duplicate logic with an equivalent
lowering pass that is used by all the back-ends.
Note: on i965, there is a slight increase in instruction count. For
example, a loop like this:
for (int i = 0; i < 100; i++) {
total += i;
}
would previously compile down to this (vec4) native code:
mov(8) g4<1>.xD 0D
mov(8) g8<1>.xD 0D
loop:
cmp.ge.f0(8) null g8<4;4,1>.xD 100D
(+f0) break(8)
add(8) g5<1>.xD g5<4;4,1>.xD g4<4;4,1>.xD
add(8) g8<1>.xD g8<4;4,1>.xD 1D
add(8) g4<1>.xD g4<4;4,1>.xD 1D
while(8) loop
After this patch, the "(+f0) break(8)" turns into:
(+f0) if(8)
break(8)
endif(8)
because the back-end isn't smart enough to recognize that "if
(condition) break;" can be done using a conditional break instruction.
However, it should be relatively easy for a future peephole
optimization to properly optimize this.
Reviewed-by: Jordan Justen <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
| |
I want to reuse them in opt_algebraic.
v2: Merge in Chris Forbes's break fix.
Reviewed-by: Jordan Justen <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
v2: Add comments on the purpose of the auxiliary data structures.
Check for atomic counter overlaps. Use the contains_atomic()
convenience method. Add static assert with the number of expected
shader stages.
v3: Don't resize atomic arrays.
v4: Add comment on the reason why we don't resize atomic counter
arrays. Use 'strcmp(...) == 0' instead of '!strcmp(...)'.
v5 (idr): Don't use STL in the linker.
Signed-off-by: Francisco Jerez <[email protected]>
Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This only operates on constant/uniform values for now, because otherwise I'd
have to deal with killing my available CSE entries when assignments happen,
and getting even this working in the tree ir was painful enough.
As is, it has the following effect in shader-db:
total instructions in shared programs: 1524077 -> 1521964 (-0.14%)
instructions in affected programs: 50629 -> 48516 (-4.17%)
GAINED: 0
LOST: 0
And, for tropics, that accounts for most of the effect, the FPS
improvement is 11.67% +/- 0.72% (n=3).
v2: Use read_only field of the variable, manually check the lod_info union
members, use get_num_operands(), rename cse_operands_visitor to
is_cse_candidate_visitor, move all is-a-candidate logic to that
function, and call it before checking for CSE on a given rvalue, more
comments, use private keyword.
Reviewed-by: Paul Berry <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
We don't actually use anything from builtin_function.cpp, so we don't
need to generate it anymore.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Matt Turner <[email protected]>
Acked-by: Paul Berry <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This creates a new replacement for the existing built-in function code.
The new module lives in builtin_functions.cpp (not builtin_function.cpp)
and exists in parallel with the existing system. It isn't used yet.
The new built-in function code takes a significantly different approach:
Instead of implementing built-ins via printed IR, build time scripts,
and run time parsing, we now implement them directly in C++, using
ir_builder. This translates to faster load times, and a much less
complex build system.
It also takes a different approach to built-in availability: each
signature now stores a boolean predicate, which makes it easy to
construct arbitrary expressions based on _mesa_glsl_parse_state's
fields. This is much more flexible than the old system, and also
easier to use.
Built-ins are also now stored in a single gl_shader object, rather
than being spread out across a number of shaders that need to be linked.
When searching for a matching prototype, we simply consult the
availability predicate. This also simplifies the code.
v2: Incorporate Matt Turner's feedback: use the new fma() function rather
than expr(). Don't expose textureQueryLOD() in GLSL 4.00 (since it
was renamed to textureQueryLod()). Also correct some #undefs.
v3: Incorporate Paul Berry's feedback: rename legacy to compatibility;
add comments to explain a few things; fix uvec availability; include
shaderobj.h instead of repeating the _mesa_new_shader prototype.
v4: Fix lack of TEX_PROJECT on textureProjGrad[Offset] (caught by oglc).
Add an out_var convenience function (more feedback by Matt Turner).
v5: Rework availability predicates for Lod functions. They were broken.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Matt Turner <[email protected]>
Enthusiastically-acked-by: Paul Berry <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
This eliminates built-in varyings such as gl_Color, gl_SecondaryColor,
gl_TexCoord, and gl_FogFragCoord if they are unused by the next stage or
not written at all (e.g. gl_TexCoord elements). The gl_TexCoord array is
broken down into separate vec4s if needed.
v2: - use a switch statement in varying_info_visitor::visit(ir_variable*)
- use snprintf
- disable the optimization for GLES2
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Over the last few years, the compiler has grown to support 7 different
language versions and 6 extensions that add new built-in types. With
more and more features being added, some of our core code has devolved
into an unmaintainable spaghetti of sorts.
A few problems with the old code:
1. Built-in types are declared...where exactly?
The types in builtin_types.h were organized in arrays by the language
version or extension they were introduced in. It's factored out to
avoid duplicates---every type only exists in one array. But that
means that sampler1D is declared in 110, sampler2D is in core types,
sampler3D is a unique global not in a list...and so on.
2. Spaghetti call-chains with weird parameters:
generate_300ES_types calls generate_130_types which calls
generate_120_types and generate_EXT_texture_array_types, which calls
generate_110_types, which calls generate_100ES_types...and more
Except that ES doesn't want 1D types, so we have a skip_1d parameter.
add_deprecated also falls into this category.
3. Missing type accessors.
Common types have convenience pointers (like glsl_type::vec4_type),
but others may not be accessible at all without a symbol table (for
example, sampler types).
4. Global variable declarations in a header file?
#include "builtin_types.h" in two C++ files would break the build.
The new code addresses these problems. All built-in types are declared
together in a single table, independent of when they were introduced.
The macro that declares a new built-in type also creates a convenience
pointer, so every type is available and it won't get out of sync.
The code to populate a symbol table with the appropriate types for a
particular language version and set of extensions is now a single
table-driven function. The table lists the type name and GL/ES versions
when it was introduced (similar to how the lexer handles reserved
words). A single loop adds types based on the language version.
Explicit extension checks then add additional types. If they were
already added based on the language version, glsl_symbol_table simply
ignores the request to add them a second time, meaning we don't need
to worry about duplicates and can simply list types where they belong.
v2: Mark uvecs and shadow samplers as ES3 only, and 1DArrayShadow as
unsupported in ES entirely. Add a touch more doxygen.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Verify that interface blocks match when combining compilation
units at the same stage. (For example, when merging all vertex
shaders.)
Fixes piglit glsl-1.50 test:
* linker/interface-blocks-multiple-vs-member-count-mismatch.shader_test
v5 (Ken): Rename to link_interface_blocks.cpp and drop the separate .h
file for consistency with other linker code. Remove "ok" variable.
Fold cross_validate_interface_blocks into its caller.
Signed-off-by: Jordan Justen <[email protected]>
Signed-off-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
| |
Convert interface blocks with instance names into flat
interface blocks without an instance name.
Signed-off-by: Jordan Justen <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This will eventually replace do_vec_index_to_cond_assign. This lowering
pass is called in all the places where do_vec_index_to_cond_assign or
do_vec_index_to_swizzle is called.
v2: Use WRITEMASK_* instead of integer literals. Use a more concise
method of generating broadcast_index. Both suggested by Eric.
v3: Use a series of scalar compares instead of a single vector compare.
Suggested by Eric and Ken. It still uses 'if (cond) v.x = y;' instead
of conditional assignments because ir_builder doesn't do conditional
assignments, and I'd rather keep the code simple.
Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This pass flips (matrix * vector) operations to (vector *
matrixTranspose) for certain built-in matrices (currently
gl_ModelViewProjectionMatrix and gl_TextureMatrix).
This is equivalent, but results in dot products rather than multiplies
and adds. On some hardware, this is more efficient.
This pass is conditionalized on ctx->mvp_with_dp4, the flag drivers set
to indicate they prefer dot products.
Improves performance in Lightsmark by 1.01131% +/- 0.162069% (n = 10)
on a Haswell GT2 system. Passes Piglit on Ivybridge.
v2: Use struct gl_shader_compiler_options instead of plumbing through
another boolean flag for this purpose.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
I love 800+ line switch-statements as much as the next guy... Future
commits will make changes to this part of the AST-to-HIR conversion, and
extracting this code will make that a bit easier.
Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
GLBenchmark 2.7's shaders contain conditional blocks like:
if (x) {
if (y) {
...
}
}
where the outer conditional's then clause contains exactly one statement
(the nested if) and there are no else clauses. This can easily be
optimized into:
if (x && y) {
...
}
This saves a few instructions in GLBenchmark 2.7:
total instructions in shared programs: 11833 -> 11649 (-1.55%)
instructions in affected programs: 8234 -> 8050 (-2.23%)
It also helps CS:GO slightly (-0.05%/-0.22%). More importantly,
however, it simplifies the control flow graph, which could enable other
optimizations.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Matt Turner <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Jordan Justen <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Also slightly change the compatibility test. Instead of comparing the
offsets of the block variables, compare the packing mode of the blocks.
Ideally we don't want to assign the offsets until a later stage of
linking.
This is put in a new file called link_uniform_blocks.cpp. Some new
functions related to uniform blocks are going to live in that file as
well.
Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Carl Worth <[email protected]>
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Lower them to arithmetic and bit manipulation expressions.
v2: Rewrite using ir_builder [for idr].
v3: Comment typos. [for mattst88]
v4: Fix arithmetic error in comments.
Factor out a shift instruction.
Don't heap allocate factory.instructions.
[for paul]
Reviewed-by: Ian Romanick <[email protected]> (v2)
Reviewed-by: Matt Tuner <[email protected]> (v3)
Reviewed-by: Paul Berry <[email protected]> (v4)
Signed-off-by: Chad Versace <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Squashed with two reverts:
Revert "android: Update for builtin_stubs.cpp move"
This reverts commit c0def90ede1e939173041b8785303de90f8fdc6c.
Revert "scons: Update for builtin_stubs.cpp"
This reverts commit 8ac4b82699ad0a59ae6ae6d3415702eaa5d4fe3b.
Tested-by: Andreas Boll <[email protected]>
Tested-on-Android-by: Chad Versace <[email protected]>
|
| |
|
|
|
|
| |
Removing the subdirectory recursion provides a small speed up.
Tested-by: Andreas Boll <[email protected]>
|
| |
|
|
|
|
|
|
| |
linker.cpp is getting pretty big, and we're about to add even more
varying packing code, so split out the linker code that concerns
varyings to its own file.
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This lowering pass generates GLSL code that manually packs varyings
into vec4 slots, for the benefit of back-ends that don't support
packed varyings natively.
No functional change--the lowering pass is not yet used.
Reviewed-by: Eric Anholt <[email protected]>
v2: Don't use ir_hierarchical_visitor--just loop over instructions
directly. Also, make the names of the packed varyings include the
names of the original varyings that were packed into them.
|
| |
|
|
|
|
|
|
| |
Fixes SCons build regression introduced with commit
a665cf1226b80ec52a0c1a4a38378df4389e8ebf.
Signed-off-by: Vinson Lee <[email protected]>
Tested-by: Vinson Lee <[email protected]>
|
| |
|
|
|
| |
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
| |
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
| |
Like in src/mesa, use GLSL_BUILDDIR/GLSL_SRCDIR to unambiguously
distinguish between in-tree and generated files.
Reviewed-by: Eric Anholt <[email protected]>
Signed-off-by: Christopher James Halse Rogers <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
| |
v2: Reduce the impenetrable code in emit_ubo_loads() by 23 lines by keeping
the ir_variable as the variable part of the offset from handle_rvalue(),
and track the constant offsets from that with a plain old integer value,
avoiding a bunch of temporary variables in the array and struct handling.
Also, fix file description doxygen.
v3: Fix a row vs col typo, and fix spelling in a comment.
Reviewed-by: Eric Anholt <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
v2: Use AM_V_GEN to silence generated code rules. Add BUILT_SOURCES to CLEANFILES
v3:
- Fix an accidental // in a path
- Use automake make rules for lex/yacc rather than writing our own
- Update .gitignore appropriately
- Build a libglcpp convenience library rather than awkwardly including
the files in libglsl and delegating the generation
- Remove libglsl.a compatibility link on clean
v4:
- Automake's rules for lex/yacc make .cc if source is .ll or .yy, and apparently we
must use those extensions "because of scons", so update everywhere glsl_parser.cpp
-> glsl_parser.cc and glsl_lexer.cpp -> glsl_lexer.cc. This fixes 'make tarballs'
and building with dricore enabled.
Signed-off-by: Jon TURNEY <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Tested-by: Matt Turner <[email protected]>
|
| | |
|
| |
|
|
|
| |
See e86c40a84d241b954594f5ae7df9b9c3fc797a4e for reasoning. In the
process I did s/:=/=/ to shut up automake about nonportable make syntax.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
v2: Fix handling of arrays-of-structure. Thanks to Eric Anholt for
pointing this out.
v3: Minor comment change based on feedback from Ken.
Fixes piglit glsl-1.20/execution/uniform-initializer/fs-structure-array
and glsl-1.20/execution/uniform-initializer/vs-structure-array.
Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
| |
Previously, I tried implementing this in the i965 driver, but did so
in a way that violated the intent of the spec, and broke Tropics.
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
| |
This conflicts with the GLSL 1.30+ rules for derivatives after a
discard has occurred.
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
The C++ constructors with placement new, while functional, are
extremely verbose, leading to generation of simple GLSL IR expressions
like (a * b + c * d) expanding to many lines of code and using lots of
temporary variables. By creating a new ir_builder.h that puts simple
generators in our namespace and taking advantage of ralloc_parent(),
we can generate much more compact code, at a minor runtime cost.
v2: Replace ir_instruction usage with just ir_rvalue.
v3: Drop remaining missed as_rvalue() in v2.
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
I've had this code laying around almost done for a long time. The
idea is like opt_structure_splitting, that we've got a bunch of
transforms at the GLSL IR level that only understand scalars and
vectors, which just skip complicated dereferences. While driver
backends may manage some optimization after they split matrices up
themselves, it would be better to bring all of our optimization to
bear on the problem.
While I wasn't expecting changes quite yet, a few programs end up
winning: a gstreamer convolution shader, and the Humus dynamic
branching demo:
Total instructions: 269430 -> 269342
3/2148 programs affected (0.1%)
1498 -> 1410 instructions in affected programs (5.9% reduction)
|
| |
|
|
|
|
|
|
| |
automake uses variables named *_SOURCES.
Reviewed-by: Eric Anholt <[email protected]>
Tested-by: Eric Anholt <[email protected]>
Signed-off-by: Matt Turner <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
This is similar to Gallium's existing glsl_to_tgsi::remove_output_read
lowering pass, but done entirely inside the GLSL compiler.
Signed-off-by: Vincent Lejeune <[email protected]>
Signed-off-by: Kenneth Graunke <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
|
| |
|
|
|
|
| |
It's only about builtins, not variables in general.
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
With the hope that Android.mk and SConscript can share the file to reduce
future breakage.
Reviewed-by: Brian Paul <[email protected]>
Reviewed-by: Chad Versace <[email protected]>
|
