Top-level instructions now get NULL as their default type (since type is
irrelevant for things like ir_function), while ir_rvalues get error_type
by default.
This should make it easier to tell if we've forgotten to set a type. It
also fixes some "Conditional jump or move depends on uninitialized
value" errors in valgrind caused by ir_validate examining the type of
top level ir_instructions, which weren't set.
Assignments can only exist at the top level instruction stream; the
residual value is handled by assigning the value to a temporary and
returning an ir_dereference_variable of that temporary.
Variables with mode ir_var_temporary were causing an out of bounds array
access and filling my screen with rubbish. I'm not sure if "temporary"
is the right thing to print.
It's really hard to believe that this case has been broken, but apparently
no test previously exercised it. So this commit adds such a test and fixes
it by making a copy of the argument token-list before expanding it.
This fix causes the following glean tests to now pass:
glsl1-Preprocessor test 6 (#if 0, #define macro)
glsl1-Preprocessor test 7 (multi-line #define)
If we put the protos in separate ir_functions, they wouldn't be found
at lookup time for linking.
Fixes:
glsl-fs-texture2d-bias
glsl-fs-texture2dproj-bias
glsl-fs-texture2dproj-bias-2
glsl-lod-bias
glsl1-texture2D(), computed coordinate
Previously, the compiler expected the result of the multiplication to
be of the same type as the vector. This is correct for square
matrices, but wrong for all others.
We fix this by instead expecting a vector with the same number of rows
as the matrix (for the case of M*v with a column vector) or the same
number of columns as the matrix (for v*M with a row vector).
This fix causes the following four glean tests to now pass:
glsl1-mat4x2 * vec4
glsl1-vec2 * mat4x2 multiply
glsl1-vec3 * mat4x3 multiply
glsl1-vec4 * mat3x4 multiply
Since GLSL permits arrays of structures, we need to store each element
as an ir_constant*, not just ir_constant_data.
Fixes parser tests const-array-01.frag, const-array-03.frag,
const-array-04.frag, const-array-05.frag, though 03 and 04 generate the
wrong code.
Implicit conversions were not being performed, nor was there any
type checking - it was possible to have, say, var->type == float
and var->constant_value->type == int. Later use of the constant
expression would trigger an assertion.
Fixes piglit test const-implicit-conversion.frag.
This is an invasive set of changes. Each user shader tracks a set of other
shaders that contain built-in functions. During compilation, function
prototypes are imported from these shaders. During linking, the
shaders are linked with these built-in-function shaders just like with
any other shader.
In both the preprocessor and in the compiler proper, we use a custom
yyltype struct to allow tracking the source-string number in addition
to line and column. However, we were previously relying on bison's
default initialization of the yyltype struct which of course is not
aware of the source field and leaves it uninitialized.
We fix this by defining our own YYLLOC_DEFAULT macro expanding on the
default version (as appears in the bison manual) and adding
initialization of yylloc.source.
Previously, any occurence of the unary plus operator would trigger a
bogus type mismatch error. Fix this by making the ast_plus case look
more like the ast_neg case as far as type-checking is concerned.
With this change the shaders/CorrectPreprocess8.frag test in piglit
now passes.
This is quite a large patch because breaking it into smaller pieces
would result in the tree being intermitently broken. The big changes
are:
* Add the ir_var_temporary variable mode
* Change the ir_variable constructor to take the mode as a
parameter and correctly specify the mode for all ir_varables.
* Change the linker to not cross validate ir_var_temporary
variables.
* Change the linker to pull all ir_var_temporary variables from
global scope into 'main'.
Since the types are singletons across the lifetime of the compiler,
repeatedly compiling a program with the same structure type defined
would drop a copy of the array on the floor per compile.
This is a bit tricky because the static GLSL types are not called with
the talloc-based new, so we have to use the global type context, which
may not be initialized yet.
We regularly do lookups on the field names of the structure to find
the types within the struct, so returning a structure type with bad
names will lead to lots of error types being found.
