| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Variables have types, expression trees have types, but statements don't.
Rather than have a nonsensical field that stays NULL in the base class,
just move it to where it makes sense.
Fix up a few places that lazily used ir_instruction even though they
actually knew the particular subclass.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Aside from ir_call, our IR is cleanly split into two classes:
- Statements (typeless; used for side effects, control flow)
- Values (deeply nestable, pure, typed expression trees)
Unfortunately, ir_call confused all this:
- For void functions, we placed ir_call directly in the instruction
stream, treating it as an untyped statement. Yet, it was a subclass
of ir_rvalue, and no other ir_rvalue could be used in this way.
- For functions with a return value, ir_call could be placed in
arbitrary expression trees. While this fit naturally with the source
language, it meant that expressions might not be pure, making it
difficult to transform and optimize them. To combat this, we always
emitted ir_call directly in the RHS of an ir_assignment, only using
a temporary variable in expression trees. Many passes relied on this
assumption; the acos and atan built-ins violated it.
This patch makes ir_call a statement (ir_instruction) rather than a
value (ir_rvalue). Non-void calls now take a ir_dereference of a
variable, and store the return value there---effectively a call and
assignment rolled into one. They cannot be embedded in expressions.
All expression trees are now pure, without exception.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, ir_call can be used as either a statement (for void
functions) or a value (for non-void functions). This is rather awkward,
as it's the only class that can be used in both forms.
A number of places use ir_call::get_error_instruction() to construct a
generic value of error_type. If ir_call is to become a statement, it
can no longer serve this purpose.
Unfortunately, none of our classes are particularly well suited for
this, and creating a new one would be rather aggrandizing. So, this
patch introduces ir_rvalue::error_value(), a static method that creates
an instance of the base class, ir_rvalue. This has the nice property
that you can't accidentally try and access uninitialized fields (as it
doesn't have any). The downside is that the base class is no longer
abstract.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
generate_call() and ast_function_expression::hir() both tried to verify
that 'out' and 'inout' parameters used l-values. Irritatingly, it
turned out that this was not redundant; both checks caught -some- cases.
This patch combines the two into a single "complete" function that does
all the parameter mode checking. It also adds a comment clarifying why
AST-level checking is necessary in the first place.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We used to have one big function, match_signature_by_name, which found
a matching signature, performed out-parameter conversions, and generated
the ir_call. As the code for matching against built-in functions became
more complicated, I split it internally, creating generate_call().
However, I left the same awkward interface. This patch splits it into
three functions:
1. match_signature_by_name()
This now takes a name, a list of parameters, the symbol table, and
returns an ir_function_signature. Simple and one purpose: matching.
2. no_matching_function_error()
Generate the "no matching function" error and list of prototypes.
This was complex enough that I felt it deserved its own function.
3. generate_call()
Do the out-parameter conversion and generate the ir_call. This
could probably use more splitting.
The caller now has a more natural workflow: find a matching signature,
then either generate an error or a call.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
Somethings, like pre-increment operations, were not previously caught.
After the 8.0 release, this code needs some major refactoring and
clean-up. It's a mess. :(
Signed-off-by: Ian Romanick <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42755
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, we would fail to compile the following shader due to a bug
in lazy built-in importing:
#version 130
void main() {
float f = abs(5.0);
int i = abs(5);
}
The first call, abs(5.0), would fail to find a local signature, look
through the built-ins, and import "float abs(float)".
The second call, abs(5), would find the newly imported float signature
in the local shader, and settle for that. Unfortunately, it failed to
search the built-ins for the correct/exact signature, "int abs(int)".
Thus, abs(5) ended up being a float, causing a bizarre type error when
we tried to assign it to an int.
Fixes piglit test builtin-overload-matching.frag.
This is /not/ a candidate for stable branches, as it should only be
possible to trigger this bug using GLSL 1.30's built-in functions that
take integer arguments. Plus, the changes are fairly invasive.
Signed-off-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
match_function_by_name performs two fairly separate tasks:
1. Hunt down the appropriate ir_function_signature for the callee.
2. Generate the actual ir_call (assuming we found the callee).
Both of these are complicated. The first has to handle exact/inexact
matches, lazy importing of built-in prototypes, different scoping rules
for 1.10, 1.20+, and ES. Not to mention printing a user-friendly error
message with pretty-printed "maybe you meant this" candidate signatures.
The second has to deal with void/non-void functions, pre-call implicit
conversions for "in" parmeters, and post-call "out" call conversions.
Trying to do both in one function is just too unwieldy. Time to split.
This patch purely moves the code to generate an ir_call into a separate
function and reindents it. Otherwise, the code is identical.
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Very simple shaders don't actually use GLSL built-ins. For example:
- gl_Position = gl_ModelViewProjectionMatrix * gl_Vertex;
- gl_FragColor = vec4(0.0);
Both of the shaders used by _mesa_meta_glsl_Clear() also qualify.
By waiting to initialize the built-ins until the first time we need to
look for a signature, we can avoid the overhead entirely in these cases.
Makes piglit run roughly 18% faster (255 vs. 312 seconds).
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
|
|
|
|
|
|
|
| |
while debugging texelFetchOffset we kept hitting the assert.
Signed-off-by: Dave Airlie <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
| |
Otherwise we continue and hit the "Illegal formal parameter mode"
assertion.
Fixes negative compile test texelFetchOffset.frag in piglit.
Signed-off-by: Kenneth Graunke <[email protected]>
Signed-off-by: Dave Airlie <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
When an out parameter undergoes an implicit type conversion, we need
to store it in a temporary, and then after the call completes, convert
the resulting value. In other words, we convert code like the
following:
void f(out int x);
float value;
f(value);
Into IR that's equivalent to this:
void f(out int x);
float value;
int out_parameter_conversion;
f(out_parameter_conversion);
value = float(out_parameter_conversion);
This transformation needs to happen during ast-to-IR convertion (as
opposed to, say, a lowering pass), because it is invalid IR for formal
and actual parameters to have types that don't match.
Fixes piglit tests
spec/glsl-1.20/compiler/qualifiers/out-conversion-int-to-float.vert and
spec/glsl-1.20/execution/qualifiers/vs-out-conversion-*.shader_test,
and bug 39651.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=39651
Reviewed-by: Chad Versace <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Rearranged the logic for converting the ast for a function call to
hir, so that we constant fold before emitting any IR. Previously we
would emit some IR, and then only later detect whether we could
constant fold. The unnecessary IR would usually get cleaned up by a
later optimization step, however in the case of a builtin function
being used to compute an array size, it was causing an assertion.
Fixes Piglit test array-size-constant-relational.vert.
Reviewed-by: Kenneth Graunke <[email protected]>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=38625
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The ast-to-hir conversion needs to emit function signatures in two
circumstances: when a function declaration (or definition) is
encountered, and when a built-in function is encountered.
To avoid emitting a function signature in an illegal place (such as
inside a function), emit_function() checked whether we were inside a
function definition, and if so, emitted the signature before the
function definition.
However, this didn't cover the case of emitting function signatures
for built-in functions when those built-in functions are called from
inside the constant integer expression that specifies the length of a
global array. This failed because when processing an array length, we
are emitting IR into a dummy exec_list (see process_array_type() in
ast_to_hir.cpp). process_array_type() later checks (via an assertion)
that no instructions were emitted to the dummy exec_list, based on the
reasonable assumption that we shouldn't need to emit instructions to
calculate the value of a constant.
This patch changes emit_function() so that it emits function
signatures at toplevel in all cases.
This partially fixes bug 38625
(https://bugs.freedesktop.org/show_bug.cgi?id=38625). The remainder
of the fix is in the patch that follows.
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Array constructors obey narrower conversion rules than other constructors
[1] --- they use the implicit conversion rules [2] instead of the scalar
constructor conversions [3]. But process_array_constructor() was
incorrectly applying the broader rules.
[1] GLSL 1.50 spec, Section 5.4.4 Array Constructors, page 52 (58 of pdf)
[2] GLSL 1.50 spec, Section 4.1.10 Implicit Conversions, page 25 (31 of pdf)
[3] GLSL 1.50 spec, Section 5.4.1 Conversion, page 48 (54 of pdf)
To fix this, first check (with glsl_type::can_be_implicitly_converted_to)
if an implicit conversion is legal before performing the conversion.
Fixes:
piglit:spec/glsl-1.20/compiler/structure-and-array-operations/array-ctor-implicit-conversion-bool-float.vert
piglit:spec/glsl-1.20/compiler/structure-and-array-operations/array-ctor-implicit-conversion-bvec*-vec*.vert
Note: This is a candidate for the 7.10 and 7.11 branches.
Reviewed-by: Kenneth Graunke <[email protected]>
Signed-off-by: Chad Versace <[email protected]>
|
|
|
|
|
|
|
|
| |
Also clarify the documentation for one of the parameters.
Reviewed-by: Paul Berry <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Inspired by a patch from Bryan Cain <[email protected]>.
Fixes piglit tests:
- ctor-int-uint.vert
- ctor-ivec4-uvec4.vert
- ctor-uint-int.vert
- ctor-uvec4-ivec4.vert
Signed-off-by: Kenneth Graunke <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
Reviewed-by: Eric Anholt <[email protected]>
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Track variables, functions, and types during parsing. Use this
information in the lexer to return the currect "type" for identifiers.
Change the handling of structure constructors. They will now show up
in the AST as constructors (instead of plain function calls).
Fixes piglit tests constructor-18.vert, constructor-19.vert, and
constructor-20.vert. Also fixes bugzilla #29926.
NOTE: This is a candidate for the 7.9 and 7.10 branches.
|
|
|
|
|
|
|
|
| |
This was my mistake when converting from talloc to ralloc. I was
confused because the other calls in the function are to asprintf_append
and the original code used str as the context rather than NULL.
Fixes bug #33823.
|
| |
|
|
|
|
|
|
|
|
| |
This annotation is for an "in" function parameter for which it is only legal
to pass constant expressions. The only known example of this, currently,
is the textureOffset functions.
This should never be used for globals.
|
| |
|
|
|
|
|
|
| |
Improves the cases when:
* an explicit assignment references the read-only variable
* an 'out' or 'inout' function parameter references the read-only variable
|
|
|
|
|
|
|
|
|
|
|
|
| |
The original lazy built-in importing patch did not add the newly created
function to the symbol table, nor actually emit it into the IR stream.
Adding it to the symbol table is non-trivial since importing occurs when
generating some ir_call in a nested scope. A new add_global_function
method, backed by new symbol_table code in the previous patch, handles
this.
Fixes bug #32030.
|
|
|
|
|
|
|
|
| |
This makes a very simple 1.30 shader go from 196k of memory to 9k.
NOTE: This -may- be a candidate for the 7.9 branch, as the benefit is
substantial. However, it's not a simple change, so it may be wiser to
wait for 7.10.
|
|
|
|
|
|
|
| |
When the semantics of write masks in assignments were changed, this
code was not correctly updated.
Fixes piglit test glsl-mat-from-vec-ctor-01.
|
| |
|
|
|
|
| |
The "instructions" variable -is- used, so the cast to void can go away.
|
|
|
|
|
| |
Fixes freedesktop.org bug #31101 as well as piglit test cases
assignment-type-mismatch.vert and constructor-28.vert.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
It turns out that most people new to this IR are surprised when an
assignment to (say) 3 components on the LHS takes 4 components on the
RHS. It also makes for quite strange IR output:
(assign (constant bool (1)) (x) (var_ref color) (swiz x (var_ref v) ))
(assign (constant bool (1)) (y) (var_ref color) (swiz yy (var_ref v) ))
(assign (constant bool (1)) (z) (var_ref color) (swiz zzz (var_ref v) ))
But even worse, even we get it wrong, as shown by this line of our
current step(float, vec4):
(assign (constant bool (1)) (w)
(var_ref t)
(expression float b2f (expression bool >=
(swiz w (var_ref x))(var_ref edge))))
where we try to assign a float to the writemasked-out x channel and
don't supply anything for the actual w channel we're writing. Drivers
right now just get lucky since ir_to_mesa spams the float value across
all the source channels of a vec4.
Instead, the RHS will now have a number of components equal to the
number of components actually being written. Hopefully this confuses
everyone less, and it also makes codegen for a scalar target simpler.
Reviewed-by: Kenneth Graunke <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
|
|
|
| |
Everything but 1.10 supports this, so just change the check to ==.
|
|
|
|
|
|
|
| |
The code for handling implicit conversions should probably get
refactored, but for now, this is easy.
Fixes piglit test constructor-26.vert.
|
|
|
|
| |
I'm not sure if this is strictly necessary, but it seems wise.
|
|
|
|
| |
Fixes piglit test constructor-27.vert.
|
|
|
|
| |
Fixes piglit test cases glsl-[fv]s-all-0[12].
|
|
|
|
|
|
|
|
|
|
|
| |
If the matrix being constructed was larger than the source matrix, it
would overwrite the lower-right part of the matrix with the wrong
values, rather than leaving it as the identity matrix.
For example, constructing a mat4 from a mat2 should only use a writemask
of "xy" when copying from the source, but was using "xyzw".
Fixes the code generated by piglit test constructor-23.vert.
|
| |
|
|
|
|
|
|
|
|
|
| |
This was triggering even for matrix-from-matrix constructors. It is
perfectly legal to construct a mat3 from a mat2 - the rest will be
filled in by the identity matrix.
Changes piglit test constructor-23.vert from FAIL to PASS, but the
generated code is incorrect.
|
|
|
|
| |
There are no integer matrix types, so switching on them is silly.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Make two passes over the constructor parameters. Write all of the
constants in a single write, then write the non-constants one at a
time. This causes the fragment shader
varying float g;
void main()
{
gl_FragColor = vec4(0.0, g, 0.0, 1.0);
}
to generate
(function main
(signature void (parameters )
(
(declare (temporary ) vec4 vec_ctor@0x8580058)
(assign (constant bool (1)) (xzw) (var_ref vec_ctor@0x8580058) (constant vec4 (0.000000 0.000000 0.000000 1.000000)) )
(assign (constant bool (1)) (y) (var_ref vec_ctor@0x8580058) (swiz xxxx (var_ref g@0x8580218) ))
(assign (constant bool (1)) (xyzw) (var_ref gl_FragColor@0x84d32a0) (var_ref vec_ctor@0x8580058) )
))
)
instead of
(function main
(signature void (parameters )
(
(declare (temporary ) vec4 vec_ctor@0x8580058)
(assign (constant bool (1)) (x) (var_ref vec_ctor@0x8580058) (constant vec4 (0.000000 0.000000 0.000000 1.000000)) )
(assign (constant bool (1)) (y) (var_ref vec_ctor@0x8580058) (swiz xxxx (var_ref g@0x8580218) ))
(assign (constant bool (1)) (z) (var_ref vec_ctor@0x8580058) (constant vec4 (0.000000 0.000000 0.000000 1.000000)) )
(assign (constant bool (1)) (w) (var_ref vec_ctor@0x8580058) (constant vec4 (0.000000 0.000000 0.000000 1.000000)) )
(assign (constant bool (1)) (xyzw) (var_ref gl_FragColor@0x84d32a0) (var_ref vec_ctor@0x8580058) )
))
)
A similar optimization could be done for matrix constructors, but it
is a little more complicate there.
|
|
|
|
|
|
|
| |
Completely initialize data that is passed to ir_constant constructor.
Fixes piglit glsl-orangebook-ch06-bump valgrind uninitialized variable
error on softpipe and llvmpipe.
|
|
|
|
| |
Make glsl include only main/core.h from core mesa.
|
|
|
|
|
|
|
|
|
| |
This error led to an assertion failure for some constructors of
non-square matrices. It only occured in matrices where the number of
columns was greater than the number of rows. It didn't even always
occur on those.
Fixes piglit glslparsertest case constructor-16.vert.
|
| |
|
| |
|
|
|
|
|
|
|
| |
Previously the in-line matrix and vector constructors would generate
swizzles in the LHS. The code is actually more clear if it just
generates the masked assignments instead of relying on the
ir_assignment constructor to convert the swizzles to write masks.
|
| |
|
|
|
|
|
| |
ir_variable always strdups the incoming name so that it matches the
lifetime of the ir_variable.
|