test1.glsl

// This appears to hang but actually just takes 50 seconds to link in glLinkProgram 32-bit
// if not in shader cache (on an old Skylake i7-6700K @ 4.00 Ghz) targeting a 980Ti
// on 32-bit Win7/8 drivers 473.04 (and an unknown previous driver version)
// Happens with compatibility or core profile, debug or non-debug profile

#version 430
layout(local_size_x = 64) in;

layout(std430, binding=4) buffer obj_format
{
   uint shape[4096][16][16]; // slow

   //uint shape[4096][16*16]; // not slow
   //uint shape[4096*16*16]; // not slow
};

void main()
{
}

test2.glsl

// This causes an error in OpenGL glLinkProgram targeting a 980Ti
// on 32-bit Win7/8 drivers 473.04 (and an unknown previous driver version).
// Happens with compatibility or core profile, debug or non-debug profile

#version 430
layout(local_size_x = 64) in;

// these variations both cause a link error in glLinkProgram 32-bit
// most consistent errors are listed below, but sometimes each gives
// the opposite error when making small changes

#if 0
   // Link error:
   //   Compute info
   //   ------------
   //   -ogles

   layout(std430, binding=5) buffer obj_desc
   {
      struct {
         uint obj[32][2];
      } objmap[512][512];
   };

#else

   // Link error:
   //   Compute info
   //   ------------
   //   (0) : fatal error C9999: out of memory - internal malloc failed

   layout(std430, binding=5) buffer obj_desc
   {
      uint obj[512][512][32][2];
   };

#endif

void main()
{
}

zmain.c

// for reference, here's the driver program, but most of its execution is hidden in the platform library;
// see compiled executables in http://nothings.org/remote/nv_compute_multidim.zip

#define STB_PLATFORM_IMPLEMENTATION
#define STBP_MATH_TYPES
#include "stb_platform.h"
#define STB_DEFINE
#include "stb.h"
typedef unsigned int uint;

void stbp_main(int argc, char **argv)
{
   stbp_api api;
   api.window.title = "test";
   api.window.size = STBP_POINT(800,600);
   api.opengl.version = STBP_GL_VERSION(4,3);
   api.opengl.compatibility = 0;//1;
   api.opengl.debug_context = 0;//2;
   api.opengl.disable_vsync = 1;
   api.sound.disable = 1;

   stbp_init(&api);

   {
      GLuint generate_program;
      char error[512];
      char *compute_shader = stb_file("test2.glsl", NULL);
      GLuint generate_shader = stbp_compile_shader(GL_COMPUTE_SHADER, &compute_shader, 1, error, sizeof(error));
      if (generate_shader == 0)
         stbp_fatal("%s", error);

      generate_program = stbp_link_program(generate_shader, 0, NULL, 0, error, sizeof(error));
      if (generate_program == 0)
         stbp_fatal("%s", error);

      glDeleteShader(generate_shader);
   }

   stbp_fatal("success");
}