do_mat_op_to_vec could be called after constant folding. The kind of optimizations this does:

Original code: vec3 N = normalize((sys_WorldMatrix * vec4( vec3( 1.0, 0.0, 0.0 ), 0 )).xyz);
Current: tmpvar_1 = normalize((sys_WorldMatrix * vec4(1.0, 0.0, 0.0, 0.0)).xyz);
With mat_op_to_vec: tmpvar_1 = normalize(sys_WorldMatrix[0].xyz);

However just running that turns all matrix ops into vectors, which both looks long and (potentially?) has some downsides.

Find a way to only do the mat_op_to_vec when less than "all" results are used?

I have a shader that, when optimised, crashes on Nexus 7.

It's caused by 'lowp' not being applied to a variable being passed to a function's out parameter.

Optimised result:

//Func decl
void FSMain (  in vec2 InTexCoord_1,  in vec4 InColour_2,  out lowp vec4 utColour_3)
//calling code:
  vec4 xlat_temp_OutColour_7; // some other stuff removed.
 FSMain (tmpvar_8, tmpvar_9, xlat_temp_OutColour_7);

I guess the result should be:

//Func decl
void FSMain (  in vec2 InTexCoord_1,  in vec4 InColour_2,  out lowp vec4 utColour_3)
//calling code:
  lowp vec4 xlat_temp_OutColour_7; // some other stuff removed.
 FSMain (tmpvar_8, tmpvar_9, xlat_temp_OutColour_7);

I tested this and my N7 seems happy with it.

Here's the shader that causes the woe:

#ifdef PROFILE_ES2
precision mediump float;
#endif // PROFILE_ES2
//
// Global variable definitions
//
uniform sampler2D _MainTex;
//
// Function declarations
//
void FSMain( in vec2 InTexCoord, in vec4 InColour, out vec4 OutColour );
//
// Function definitions
//
void FSMain( in vec2 InTexCoord, in vec4 InColour, out vec4 OutColour ) {
    OutColour = texture2D( _MainTex, InTexCoord);
    OutColour *= InColour;
}
//
// User varying
//
varying vec4 xlat_varying_TEXCOORD0;
varying vec4 xlat_varying_COLOR0;
//
// Translator's entry point
//
void main() {
    vec4 xlat_temp_OutColour;
    FSMain( vec2(xlat_varying_TEXCOORD0), vec4(xlat_varying_COLOR0), xlat_temp_OutColour);
    gl_FragData[0] = vec4( xlat_temp_OutColour);
;
}

Al.

We build the glsl optimizer source directly into an ios app and (perhaps not the best practice) create a new glslopt_ctx for each shader we compile at runtime. Recently (possibly as of Xcode 6) we seem to be dying sometimes inside glslopt_cleanup with memory errors, or glslopt_optimize starts failing with a lot of error messages about seemingly correct code.

Before I do a lot of digging to try to narrow down a test case I wondering if creating multiple glslopt contexts per process/GL context/whatever is even safe to do, or if there is global state being stored somewhere.

Hi there,

I have trouble with compiling on Gentoo, I will probably do a fork and just fix this but also want to share.

[ 98%] Building CXX object CMakeFiles/glsl_compiler.dir/src/glsl/main.cpp.o
/usr/bin/x86_64-pc-linux-gnu-g++ -DNDEBUG -O2 -pipe -march=corei7-avx -fno-omit-frame-pointer -Os -DNDEBUG -I/tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999/include -I/tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999/src/mesa -I/tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999/src/mapi -I/tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999/src/glsl -I/tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999/src -o CMakeFiles/glsl_compiler.dir/src/glsl/main.cpp.o -c /tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999/src/glsl/main.cpp
Linking CXX executable glsl_compiler
/usr/bin/cmake -E cmake_link_script CMakeFiles/glsl_compiler.dir/link.txt --verbose=1
/usr/bin/x86_64-pc-linux-gnu-g++ -O2 -pipe -march=corei7-avx -fno-omit-frame-pointer -Os -DNDEBUG CMakeFiles/glsl_compiler.dir/src/glsl/main.cpp.o -o glsl_compiler -rdynamic libglsl_optimizer.a libglcpp-library.a libmesa.a
libglsl_optimizer.a(standalone_scaffolding.cpp.o): In function _mesa_error_no_memory': standalone_scaffolding.cpp:(.text+0xf): multiple definition of_mesa_error_no_memory'
CMakeFiles/glsl_compiler.dir/src/glsl/main.cpp.o:main.cpp:(.text+0x0): first defined here
collect2: error: ld returned 1 exit status
make[2]: *** [glsl_compiler] Error 1
make[2]: Leaving directory /tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999_build' make[1]: *** [CMakeFiles/glsl_compiler.dir/all] Error 2 make[1]: Leaving directory/tmp/portage/gfx-tools/glsl-optimizer-9999/work/glsl-optimizer-9999_build'
make: *** [all] Error 2

Looks like the _mesa_error_no_memory function is defined in
src/glsl/main.cpp and in
src/glsl/standalone_scaffolding.cpp
and that causes an error while linking.

Regards,
Iegor

Hi,
I'm having problems linking the static lib in a project.
I've added the glsl_optimizer_lib project as a reference, and the library builds fine. But then when I use
glslopt_ctx* ctx = glslopt_initialize(kGlslTargetOpenGLES20);

I get:
1>MainComponent.obj : error LNK2019: unresolved external symbol "class glslopt_ctx * __cdecl glslopt_initialize(enum glslopt_target)" (?glslopt_initialize@@YAPAVglslopt_ctx@@W4glslopt_target@@@z) referenced in function "public: __thiscall MainContentComponent::MainContentComponent(void)" (??0MainContentComponent@@QAE@XZ)

I've tried every possible way of adding the library: project reference, additional dependencies, etc.
I'm fairly sure that the compiler is finding the lib, but still. Is there something I'm missing? An extra dependency?
I've tried including the source directly but it won't build due to typedef clashes (my project also uses OpenGL) and before digging into that I thought I'd ask here.
Any help at all would be most appreciated.

Rob

I have compile error under Debian squize 64-bit:
[ 96%] Building C object CMakeFiles/glcpp.dir/src/glsl/glcpp/glcpp.c.o
In file included from /home/alex/src/glsl-optimizer-master/src/glsl/glcpp/glcpp.c:29:
/home/alex/src/glsl-optimizer-master/src/glsl/glcpp/../standalone_scaffolding.h:36: error: expected identifier or ‘(’ before string constant
/home/alex/src/glsl-optimizer-master/src/glsl/glcpp/../standalone_scaffolding.h:40: error: expected identifier or ‘(’ before string constant
/home/alex/src/glsl-optimizer-master/src/glsl/glcpp/../standalone_scaffolding.h:43: error: expected identifier or ‘(’ before string constant
make[2]: *** [CMakeFiles/glcpp.dir/src/glsl/glcpp/glcpp.c.o] Error 1
make[1]: *** [CMakeFiles/glcpp.dir/all] Error 2
make: *** [all] Error 2

if i try to enclose definitions in standalone_scaffolding.h like this:

ifdef _cplusplus

extern "C" {

endif

void _mesa_reference_shader(struct gl_context _ctx, struct gl_shader *_ptr,
struct gl_shader *sh);

ifdef _cplusplus

}

endif

i have another compilie error later:
[ 97%] Building CXX object CMakeFiles/glsl_compiler.dir/src/glsl/main.cpp.o
Linking CXX executable glsl_compiler
libglsl_optimizer.a(builtin_function.cpp.o): In function read_builtins(unsigned int, char const*, char const**, unsigned int)': builtin_function.cpp:(.text+0x74): undefined reference to_mesa_new_shader'
libglsl_optimizer.a(linker.cpp.o): In function link_shaders(gl_context*, gl_shader_program*)': linker.cpp:(.text+0x1dcb): undefined reference to_mesa_reference_shader'
linker.cpp:(.text+0x1e17): undefined reference to `_mesa_reference_shader'
collect2: ld returned 1 exit status
make[2]: *** [glsl_compiler] Error 1
make[1]: *** [CMakeFiles/glsl_compiler.dir/all] Error 2
make: *** [all] Error 2

From email: "When optimizing fragment shaders it doesn't really handle default precision declaration (it is not present in output shader source which then fails to compile)."

On OSX 10.10.1 I get the following build error trying to install the current HEAD of master:

$ npm install git+https://github.com/aras-p/glsl-optimizer.git
|
> [email protected] install ./node_modules/glsl-optimizer
> node-gyp rebuild

  CC(target) Release/obj.target/glsl_optimizer_lib/src/glsl/glcpp/glcpp-lex.o
In file included from src/glsl/glcpp/glcpp-lex.l:29:
../src/glsl/glcpp/glcpp.h:32:10: fatal error: 'util/ralloc.h' file not found
#include "util/ralloc.h"
         ^
1 error generated.
make: *** [Release/obj.target/glsl_optimizer_lib/src/glsl/glcpp/glcpp-lex.o] Error 1

Why are all opengl context constants hardcoded?
Would be nice to set them after construction of the glsl context.

CMake Error at CMakeLists.txt:21 (list):
list sub-command REMOVE_ITEM requires two or more arguments.

-- Configuring incomplete, errors occurred!

Linux chris-LIFEBOOK-AH532 3.8.0-21-generic #32-Ubuntu SMP Tue May 14 22:16:46 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux

Hi
This code have'h bugs, but not compilate:

mat4 mat;
mat[0][1] = mat[0][2] = mat[0][3] = 0.0;

with this brunch - https://github.com/aras-p/glsl-optimizer/tree/bbatkin

Hi,

Here is the error I receive at the very beginning:

In file included from src/glsl/glcpp/glcpp-parse.y:31:0:
/home/noob/CoolCode/glsl-optimizer/src/glsl/glcpp/glcpp.h:32:25: fatal error: util/ralloc.h: No such file or directory
 #include "util/ralloc.h"
                         ^
compilation terminated.
CMakeFiles/glcpp-library.dir/build.make:54: recipe for target 'CMakeFiles/glcpp-library.dir/src/glsl/glcpp/glcpp-parse.c.o' failed
make[2]: *** [CMakeFiles/glcpp-library.dir/src/glsl/glcpp/glcpp-parse.c.o] Error 1
CMakeFiles/Makefile2:60: recipe for target 'CMakeFiles/glcpp-library.dir/all' failed
make[1]: *** [CMakeFiles/glcpp-library.dir/all] Error 2
Makefile:76: recipe for target 'all' failed
make: *** [all] Error 2

Tried with CMake in both a separate build directory, and an in-source build; neither worked. Running Gentoo.

Hi,

I have something like that in my shader :

    bool InRect( vec2 uv ){
        return all( greaterThan( uv , vec2(1.0/1024.0) ) && lessThan( uv , vec2(1023.0/1024.0) ) );
    }

greaterThan and lessThan returns bvec2. So ( bvec2 && bvec2 ) return bvec2 too. Thats why I used all( bvec2 ).

glsl-optimizer sees ( bvec2 && bvec2 ) as regular bool && bool so the return type is bool. And all() doesn't allow to bool as an input, so the whole shader optimalization is broken.

BTW, is my solution enought good for testing for uv beeing inside rectangle ?

I am getting following errors while using the optimised shader.

ERROR: 0:369: Invalid call of undeclared identifier 'csel_TODO'
WARNING: 0:478: Overflow in implicit constant conversion, minimum range for lowp float is (-2,2)
WARNING: 0:650: Overflow in implicit constant convers

The test app itself generates error when trying to compile the optimised shaders.
How can i fix the errors about 'csel_TODO'?

This vertex shader:

#version 100
struct shadow_map_t
{
    mat4 transform;
};

uniform shadow_map_t ShadowMaps[1];
attribute vec3 _in_position;

void main()
{   
    gl_Position = ShadowMaps[0].transform * vec4(_in_position,1);
}

Turns into this after optimizing:

#version 100
attribute vec3 _in_position;
uniform shadow_map_t ShadowMaps[1];
void main ()
{
  vec4 tmpvar_1;
  tmpvar_1.w = 1.0;
  tmpvar_1.xyz = _in_position;
  gl_Position = (ShadowMaps[0].transform * tmpvar_1);
}

Notice the definition of the shadow_map_t struct has been lost. This was originally reproduced with a non-ES context and #version 130, but it happens for #version 100 ES as well.

EDIT:

It looks like not having an explicit reference to a variable of type "shadow_map_t" is the cause. If I change the source to:

shadow_map_t sm = ShadowMaps[0];
gl_Position = sm.transform * vec4(_in_position,1);

Then the shadow_map_t definition is preserved.

We have a case where a constant variable in an operation is being replaced with it's literal representation, dropping the precision qualifier. This variable was the only highp value in the operation, which means that the whole operation is now performed at mediump.

Here's the case (note that gl_FragCoord is mediump):

highp float time = 1.0;
highp float dither = dot( vec2( 171.0, 231.0 ), gl_FragCoord.xy + vec2( time ) );

is optimized to:

highp float tmpvar_17;
tmpvar_17 = dot (vec2(171.0, 231.0), (gl_FragCoord.xy + vec2(1.0, 1.0)));

The tmpvar_17 calculation now contains only one variable with precision, mediump gl_FragCoord, so the calculation is performed at mediump. Replacing the 1.0 literals with a highp float initialized to 1.0 fixes the issue. The bug here seems to be that constants are being replaced by literals without considering precision.

We're finding that for a fragment shader with a struct uniform, and a top-level default precision specifier, the optimizer output has the struct definition before the precision statement, leaving the precision of the struct fields undefined.

For instance given this input:

// #version 100
precision mediump float;

struct HColor {
  vec4 diffuse;
  vec4 specular;
};

uniform HColor MaterialColor;

void main ()
{
    gl_FragData[0] = MaterialColor.diffuse;
}

The optimizer produces this output:

struct HColor {
  vec4 diffuse;
  vec4 specular;
};
precision mediump float;
uniform HColor MaterialColor;
void main ()
{
  gl_FragData[0] = MaterialColor.diffuse;
}

Which does not compile.

sqrt(dot(v, v)) is slower than length(v) on Mali GPUs. (Confirmed with their offline binary compiler.)

How does one tweak the glsl optimizer to not replace certain functions? If you could point me at the relevant code, that would be awesome.

Input:

#version 130

uniform sampler2DArray tilemap;

void main() {

        vec4 col = textureLod(tilemap, vec3(0.0), 0.0);

        gl_FragColor = col;
}

Output:

#version 130
uniform sampler2DArray tilemap;
void main ()
{
  gl_FragColor = textureProjLod (tilemap, vec3(0.0, 0.0, 0.0), 0.0);
}

It's obviously wrong, and won't even compile.

When I try to build master on Ubuntu 12/04, I get the following error:

behdad:glsl-optimizer 2 (master)$ make
[ 3%] Built target glcpp-library
[ 7%] Built target mesa
[ 97%] Built target glsl_optimizer
[ 98%] Built target glsl_compiler
[ 99%] Built target glsl_test
[100%] Building CXX object CMakeFiles/glslopt.dir/contrib/glslopt/Main.cpp.o
cc1plus: warning: /home/behdad/src/github/glsl-optimizer/src/mapi: No such file or directory [enabled by default]
/home/behdad/src/github/glsl-optimizer/contrib/glslopt/Main.cpp: In function ‘bool init()’:
/home/behdad/src/github/glsl-optimizer/contrib/glslopt/Main.cpp:20:37: error: cannot convert ‘bool’ to ‘glslopt_target’ for argument ‘1’ to ‘glslopt_ctx* glslopt_initialize(glslopt_target)’
compilation terminated due to -Wfatal-errors.
make[2]: *** [CMakeFiles/glslopt.dir/contrib/glslopt/Main.cpp.o] Error 1
make[1]: *** [CMakeFiles/glslopt.dir/all

-fPIC is basically a requirement for x64 libs. Consider adding it by default? Otherwise many users will repeat this headache.

Running a rather large shader through glsl-optimizer with target ES20 gives me the following error in WebGL;

gl.getShaderInfoLog() ERROR: 0:80: 'q__19' : identifiers containing two consecutive underscores (__) are reserved as possible future keywords

This is defined in the spec, see 3.3:
http://www.opengl.org/registry/doc/GLSLangSpec.Full.1.20.8.pdf

Would be cool to add https://www.npmjs.org/package/node-pre-gyp to avoid having to rebuild the module every time, e.g. when it is installed on Travis.

Hi
error when use - gl_FragData[2] because index need < 2

but it is wrong, index 2 and 3 is valid and work perfect on devices who support ES 3.0

I'm getting this compiler error:

/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp: In function ‘void mark(gl_program*, ir_variable*, int, int, bool)’:
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:87:7: error: ‘GLbitfield64’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:87:20: error: expected ‘;’ before ‘bitfield’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:89:9: error: ‘struct gl_program’ has no member named ‘InputsRead’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:89:23: error: ‘bitfield’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:91:13: error: ‘gl_fragment_program’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:91:34: error: ‘fprog’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:91:64: error: expected primary-expression before ‘)’ token
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:91:66: error: expected ‘;’ before ‘prog’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:98:16: error: ‘struct gl_program’ has no member named ‘SystemValuesRead’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:98:36: error: ‘bitfield’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:100:9: error: ‘struct gl_program’ has no member named ‘OutputsWritten’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:100:27: error: ‘bitfield’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp: In member function ‘virtual ir_visitor_status ir_set_program_inouts_visitor::visit_enter(ir_expression*)’:
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:178:7: error: ‘gl_fragment_program’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:178:28: error: ‘fprog’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:178:58: error: expected primary-expression before ‘)’ token
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:178:60: error: expected ‘;’ before ‘prog’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp: In member function ‘virtual ir_visitor_status ir_set_program_inouts_visitor::visit_enter(ir_discard*)’:
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:190:4: error: ‘gl_fragment_program’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:190:25: error: ‘fprog’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:190:55: error: expected primary-expression before ‘)’ token
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:190:57: error: expected ‘;’ before ‘prog’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp: In function ‘void do_set_program_inouts(exec_list*, gl_program*, bool)’:
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:202:10: error: ‘struct gl_program’ has no member named ‘InputsRead’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:203:10: error: ‘struct gl_program’ has no member named ‘OutputsWritten’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:204:10: error: ‘struct gl_program’ has no member named ‘SystemValuesRead’
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:206:7: error: ‘gl_fragment_program’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:206:28: error: ‘fprog’ was not declared in this scope
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:206:58: error: expected primary-expression before ‘)’ token
/home/glsl-optimizer/src/glsl/ir_set_program_inouts.cpp:206:60: error: expected ‘;’ before ‘prog’
make[2]: *** [CMakeFiles/glsl_optimizer.dir/src/glsl/ir_set_program_inouts.cpp.o] Error 1
make[1]: *** [CMakeFiles/glsl_optimizer.dir/all] Error 2
make: *** [all] Error 2

I'm running on Ubuntu 12.04 LTS, 64-bit.

After migration from ndk-r9d to ndk-r10d I have some link errors:

src/glsl/glsl_lexer.cpp:3167: error: undefined reference to '__srget'
src/glsl/glcpp/glcpp-lex.c:1887: error: undefined reference to '__srget'

Hi aras,
I found the optimizer cannot correctly dealing with nested structure.
e.g, the following code:

struct fragInput {
highp vec4 baseTC;
}
struct fragPass {
fragInput IN;
}

will be interpreted by the optimizer as:

fragInput tmpvar_1;
tmpvar_1 = fragInput(vec4(0.0, 0.0, 0.0, 0.0));

but, struct fragInput's declaration is missing. How to solve this problem?

thanks a lot.

During fragment shader optimization it can be incorrent function call inserted such as "vector_insert_TODO".

Noticing something odd with for-loops when passing through output from hlsl2glsl. The optimised output doesn't seem to be assigning the initial value of the index at all. Noticed this as I was having glDrawArrays lock up on shaders using for loops.

For state in question (in source HLSL):

    for ( int index = 0 ; index < sampleNum ; index += 4 ) 

In GLSL after running through HLSL2GLSL

    int index = 0;
    for ( ; (index < sampleNum); index += 4) {
    ...
    }

Raw glsloptimiser output:

  tmpvar_254 = 0;
  index_230 = tmpvar_254;
  while (true) {
    ...
    if (!((index_230 < sampleNum_226))) {
      break;
    };
   ...
    int tmpvar_322;
    tmpvar_322 = (index_230 + 4);
    index_230 = tmpvar_322;
    }

Optimised glsloptimiser output:

  for (int index_5; index_5 < sampleNum_2; index_5 += 4) {
  ...
  }

The initial value is completely lost. Changing this to a non-zero value yields the same results, initially I thought it may just be an assumption on the optimisers part, but it seems to be more than that.

The problem is that the precision that the optimizer sticks to the temporary variables is causing the draw call to fail sometimes, with error 0x502.

This is a snippet from the shader without optimization:

precision mediump int;
precision mediump float;
uniform float normal_mult;
varying vec3 vertex_interp;
varying vec3 normal_interp;
attribute highp vec4 vertex_attrib;
attribute vec3 normal_attrib;
uniform highp mat4 world_transform;
uniform highp mat4 camera_inverse_transform;
void main ()
{
    highp mat4 modelview = (camera_inverse_transform * world_transform);
    highp vec4 vertex_in = vertex_attrib;
    vertex_interp = (modelview * vertex_in).xyz;
    vec3 normal_in = normal_attrib;
    normal_in*=normal_mult;
    normal_interp = normalize((modelview * vec4(normal_in,0.0)).xyz);
}

Which is optimized into this:

precision mediump float;
precision mediump int;
uniform float normal_mult;
varying highp vec3 vertex_interp;
varying highp vec3 normal_interp;
attribute highp vec4 vertex_attrib;
attribute vec3 normal_attrib;
uniform highp mat4 world_transform;
uniform highp mat4 camera_inverse_transform;
void main ()
{
    highp mat4 tmpvar_1;
    tmpvar_1 = (camera_inverse_transform * world_transform);
    vertex_interp = (tmpvar_1 * vertex_attrib).xyz;
    vec4 tmpvar_2;
    tmpvar_2.w = 0.0;
    tmpvar_2.xyz = (normal_attrib * normal_mult);
    normal_interp = normalize((tmpvar_1 * tmpvar_2).xyz);
}

If I use the optimized shader as is, the call to glDrawElements fails with error 0x502 (a.k.a. invalid operation)

If I set tmpvar_2.w to a value other than zero, or change the declaration of tmpvar_2 to highp vec4 tmpvar_2; the shader works ok.

The optimized shader is causing troubles on the iPod touch 5g running iOS 7.1.2, although it's running ok on the simulator.

I have to modify programmatically the optimized shader after the optimization, so my question is whether as a rule of thumb it's ok to propagate the precision of the variable holding the result to those temporary variables involved in the calculation.

For instance since normal_interp have highp, both tmpvar_1 and tmpvar_2 should be highp too, since those are temporary variables.

When I build and try to run "glsl_test tests/" or glsl_compiler on my own shader, the thing keeps working and never finishes.

Sample:

$ ./glsl_test tests/

** running vertex tests for OpenGL ES 2.0...

top shows that glsl_test is consuming 100% CPU, but never returns. Memory usage is stable. Same with glsl_compiler.

(I am wandering through the print visitors at this moment to find root cause / possible action, but figured I would submit an issue in case I am missing something obvious.)

Observed: The output stage (post-optimization) does not appear to emit GLSL uniform buffer objects that appear to be properly tracked during optimization.

E.g.:, a vertex shader that starts with:

version 300 es

layout (std140) uniform ModelViewProjectionConstantBuffer
{
mat4 model;
mat4 view;
mat4 projection;
} mvp;

. . . with requisite access to mvp.model, mvp.view, mvp.projection in the shader body . . .
. . . results in a shader that begins with:

version 300 es

uniform ModelViewProjectionConstantBuffer mvp;

. . . losing the layout definition entirely, resulting in errors if the resulting shader is compiled. The declaration is no longer valid.

The ir appears to track the definition okay (base type GLSL_TYPE_INTERFACE, name "ModelViewProjectionConstantBuffer", length of 3, first entry in the "fields" union is a seemingly valid representation of a mat4 named "model".

I suspect that this is an issue with the print-visiting only, will dig into it further.

src\glsl\opt_vector_splitting.cpp : 274
loop_state* ls = analyze_loop_variables(instructions);

object "ls" is created and never deleted

When rendering to multiple render targets in ES 3.0, all output locations must be explicitly specified in the fragment shader (since there is nothing like glBindFragDataLocation to control it from the application code). These qualifiers appear to be getting stripped out by the optimizer though.

For example this shader:

#version 300 es
precision mediump float;

layout(location=0) out vec4 final_color0;
layout(location=1) out vec4 final_color1;
layout(location=2) out vec4 final_color2;

void main()
{
    final_color0 = vec4(1,0,0,1);
    final_color1 = vec4(0,1,0,1);
    final_color2 = vec4(0,0,1,1);
}

Turns into this after optimizing:

#version 300
precision mediump float;
out vec4 final_color0;
out vec4 final_color1;
out vec4 final_color2;
void main ()
{
    final_color0 = vec4(1.0, 0.0, 0.0, 1.0);
    final_color1 = vec4(0.0, 1.0, 0.0, 1.0);
    final_color2 = vec4(0.0, 0.0, 1.0, 1.0);
}

Where the layout locations are now missing. The #version string also lost the "es" qualifier but that already been reported in issue 34.

The errors:

"./builtincompiler --glsl-es OilPaint.frag
Info log for OilPaint.frag:
0:0(0): error: no matching function for call to texture2D(sampler2D, vec2)' 0:0(0): error: no matching function for call tomin(float, float)'"

The fragment shader code:

//
// Shader.fsh
// Test
//
// Created by Ofer Rubinshtein on 12/2/12.
// Copyright (c) 2012 Ofer Rubinshtein. All rights reserved.
//

varying highp vec2 texOut1;
uniform sampler2D baseTexture;
uniform sampler2D overlayTexture1;
uniform highp float alpha;
uniform highp float radius;
uniform highp vec2 pixelRatio;
//uniform int step;

void main()
{
int rx = 2;
int ry = 2;
highp float lumCount[13];
highp float averageR[13];
highp float averageG[13];
highp float averageB[13];
for (int i=0; i<13; i++)
{
lumCount[i] = 0.;
averageR[i] = 0.;
averageG[i] = 0.;
averageB[i] = 0.;
}

// highp float tempMax = 0.0;
for (int i=0; i<(2_ry+1); i++)
{
int y =i;
y-=ry;
for (int j=0; j<(2_rx+1); j++)
{
int x =j;
x-=rx;
//highp float b = max(1.0-(sqrt(float(x_x+y_y))/radius), 0.0);
// highp float b = 1.0/(1.0+float(rx_rx+ry_ry));//float(r+1-abs(x))/step;
highp float b = 1.0;
highp vec2 delta = vec2(float(x)_pixelRatio.x_radius, float(y)_pixelRatio.y_radius);
highp vec2 uv = texOut1+delta;
highp vec4 rgb = texture2D(baseTexture, uv);
highp float intensity = min(((rgb.r+rgb.g+rgb.b)/3.0)_13.0, 12.0);
int curIntensity = int(intensity);
lumCount[curIntensity]+=b;
averageR[curIntensity] += rgb.r;//_b;
averageG[curIntensity] += rgb.g;//_b;
averageB[curIntensity] += rgb.b;//_b;
}
}
int maxIndex = 0;
highp float curMax;
curMax = 0.0;
for (int i=0; i<13; i++)
if (lumCount[i] > curMax)
{
curMax = lumCount[i];
maxIndex = i;
}
gl_FragColor = vec4(averageR[maxIndex] / curMax, averageG[maxIndex] / curMax, averageB[maxIndex] / curMax, 1.0);
}

Any idea what I am doing wrong? I couldn't find a lot of documentation about it.

Error:

e:\glsl-opt\src\glsl\builtin_functions.cpp(2668): error C2065: 'M_PI_2': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\builtin_functions.cpp(2670): error C2065: 'M_PI_2': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\builtin_functions.cpp(2672): error C2065: 'M_PI_4': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\builtin_functions.cpp(2715): error C2065: 'M_PI_2': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\builtin_functions.cpp(2754): error C2065: 'M_PI_2': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\builtin_functions.cpp(2809): error C2065: 'M_PI_2': nichtdeklarierter Bezeichner

Fix:

diff --git a/src/mesa/main/compiler.h b/src/mesa/main/compiler.h
index 34671dc..ce85e8f 100644
--- a/src/mesa/main/compiler.h
+++ b/src/mesa/main/compiler.h
@@ -247,6 +247,14 @@ static inline GLuint CPU_TO_LE32(GLuint x)
 #define M_PI (3.14159265358979323846)
 #endif

+#ifndef M_PI_2
+#define M_PI_2 (1.57079632679489661923)
+#endif
+
+#ifndef M_PI_4
+#define M_PI_4 (0.785398163397448309616)
+#endif
+
 #ifndef M_E
 #define M_E (2.7182818284590452354)
 #endif

Error:

e:\glsl-opt\src\glsl\ir_function.cpp(331): error C2065: 'func': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\link_atomics.cpp(65): error C2065: 'func': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\link_uniform_blocks.cpp(189): error C2065: 'func': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\link_varyings.cpp(323): error C2065: 'func': nichtdeklarierter Bezeichner
e:\glsl-opt\src\glsl\linker.cpp(1735): error C2065: 'func': nichtdeklarierter Bezeichner

Fix:

diff --git a/include/c99_compat.h b/include/c99_compat.h
index 429c601..e03e9c1 100644
--- a/include/c99_compat.h
+++ b/include/c99_compat.h
@@ -108,6 +108,10 @@
 /*
  * C99 __func__ macro
  */
+#if defined(_MSC_VER) && !defined(__func__)
+#  define __func__ __FUNCTION__
+#endif
+
 #ifndef __func__
 #  if (__STDC_VERSION__ >= 199901L)
      /* C99 */
diff --git a/src/mesa/main/errors.h b/src/mesa/main/errors.h
index 2cb7505..54d364d 100644
--- a/src/mesa/main/errors.h
+++ b/src/mesa/main/errors.h
@@ -34,7 +34,7 @@

 #ifndef ERRORS_H
 #define ERRORS_H
-
+#include "c99_compat.h"


 #ifdef __cplusplus

Being able to optimize individual shaders is great, but it would be even better if a fragment and vertex shader could be optimized at the same time - this would allow, for example, removing vertex shader code that writes to a varying that's not read by the fragment shader.

I don't know if Mesa's shader compiler/linker does this at all, but if so it'd be great to have that functionality exposed in glsl-optimizer.

The optimized output from a case statement is missing an 'if' and sequential assignments have commas missing.

Here is a typical output for a case:

bool tmpvar_3;
tmpvar_3 = bool(0);
bool tmpvar_4;
tmpvar_4 = bool(0);
(918 == style) tmpvar_3 = bool(1);
tmpvar_4 tmpvar_3 = bool(0);
if (tmpvar_3) {
float angle_5;
angle_5 = -(rotation_radians);
float tmpvar_6;
tmpvar_6 = cos(angle_5);
float tmpvar_7;
tmpvar_7 = sin(angle_5);
vec2 tmpvar_8;
tmpvar_8.x = tmpvar_6;
tmpvar_8.y = tmpvar_7;
vec2 tmpvar_9;
tmpvar_9.x = -(tmpvar_7);
tmpvar_9.y = tmpvar_6;
mat2 tmpvar_10;
tmpvar_10[0u] = tmpvar_8;
tmpvar_10[1u] = tmpvar_9;
vec3 tmpvar_11;
tmpvar_11.xy = (normal_2.xy * tmpvar_10);
tmpvar_11.z = normal_2.z;
vec3 tmpvar_12;
tmpvar_12 = normalize(diffuse_direction);
vec3 I_13;
I_13 = -(tmpvar_12);
color_1.xyz = (((ads_lighting.x * input_color.xyz) + (
(ads_lighting.y * max (dot (tmpvar_11, tmpvar_12), 0.0))
* input_color.xyz)) + ((ads_lighting.z *
pow (max (0.0, (I_13 - (2.0 *
(dot (tmpvar_11, I_13) * tmpvar_11)
)).z), 128.0)
) * vec3(1.0, 0.945, 0.878)));
tmpvar_4 = bool(1);
};

It appears that when optimizing a glsl shader that has a struct uniform, the metal output ends up with the definition of the struct appearing after the definition of xlatMtlShaderUniform which references the, at that point undefined, struct type.

For example input like this:

struct MyStruct
{
    vec4 v1;
    vec4 v2;
};

uniform MyStruct s1;

Ends up producing output like this:

struct xlatMtlShaderUniform {
  MyStruct s1;
};
;
struct MyStruct {
  float4 v1;
  float4 v2;
};

Full test shader is here https://gist.github.com/eodabash/28a01d147cbc496d14ad.

I found a couple problems with optimization of shaders that make use of OES_EGL_image_external functionality. The first is that "#extension GL_OES_EGL_image_external : enable" statements don't get printed back in the optimizer output. The second is that texture sampling calls that use a samplerExternalOES sampler seem to get converted to "textureExternal" calls which as far as I can tell is not the correct way to sample from such a sampler (the spec refers to this being an obsolete syntax that was dropped in favor of the existing sampling functions like texture2D).

A test input shader can be found here:
https://www.dropbox.com/s/vggjoa1p33uel5n/imageExternal-inES.txt?dl=0

Steps to reproduce:

• Clone repo
• Run npm install to trigger build (will call node-gyp rebuild internally)
• Run build/Release/glslOptimizer.node
• Segfault!

Fiddling around with gdb I'm able to get:

(gdb) run
Starting program: glsl-optimizer/build/Release/obj.target/glslOptimizer.node 

Program received signal SIGSEGV, Segmentation fault.
_mesa_glsl_parse_state::has_uniform_buffer_objects (this=0x0) at ../src/glsl/glsl_parser_extras.h:197
197           return ARB_uniform_buffer_object_enable || is_version(140, 300);
(gdb) quit

On a related note, #92 is reproducible on my machine (Debian 8, quadcore i5, linux kernel v4.0.0-1-amd64) . I somehow suspect a bad build via cmake might be related to a segfault-y build via gyp.

Hello,

My build failed with the error:

aras-p-glsl-optimizer-f2217b0/tests/glsl_optimizer_tests.cpp:250:22: error: ‘unlink’ was not declared in this scope
make[2]: *** [CMakeFiles/glsl_test.dir/tests/glsl_optimizer_tests.cpp.o] Error 1
make[1]: *** [CMakeFiles/glsl_test.dir/all] Error 2
make: *** [all] Error 2

Fix: add unistd.h to /tests/glsl_optimizer_tests.cpp for non-windows builds

http://gcc.gnu.org/gcc-4.7/porting_to.html

Cheers

It could be added an option to pack/unpack varyings to be able to surpass the 8 varyings limitation in some devices. For instance four floats into one vec4, two vec2 into one vec4, and so on.

The remove_unlinked_functions function causes a segmentation fault.

If you use atan, it is replaced by an if statement and other slow looking arithmetic. How can we get atan to be passed through?

After optimizing, sometimes multiple expressions are placed into a for loop. This causes the shader to crash on WebGL.

for (int i=0; i<10; i++, otherVar++) { .. }

Tested on Chrome and FireFox, OSX 10.8.5. See here:

http://glsl.heroku.com/e#17544.1

When using

float texture(samplerCubeShadow, vec4)

the glsl optimizer exchanges the function with textureProj, which does not support cube textures.

Hi Aras,

What is the current status of ES 3.0 support? I see there's a recent commit mentioning it but after getting the latest changes all our "#version 300 es" lines seem to get turned into just "#version 300" which doesn't compile.

I can submit a test case, but I'm wondering if you even intended this to be ready yet, since the "add initial OpenGL ES 3.0 support" commit doesn't look like it touched a lot of code.

Thanks,

Evan

i take version from master: https://github.com/aras-p/glsl-optimizer

set version300 es
he show message - not support 3.0, support only 1.0
bun on master page - GLSL ES versions 1.00 and 3.00 are supported.

need make brach with this ? https://github.com/aras-p/glsl-optimizer/tree/bbatkin

Or ?

When defining "invariant gl_Position" in shader, it not preserved in the optimized shader