void
piglit_init(int argc, char **argv)
{
GLuint pipe = 0;
unsigned glsl_version;
piglit_require_vertex_shader();
piglit_require_fragment_shader();
piglit_require_extension("GL_ARB_separate_shader_objects");
piglit_require_extension("GL_ARB_explicit_attrib_location");
glsl_version = pick_a_glsl_version();
vs = generate_program(vs_template, glsl_version, GL_VERTEX_SHADER,
&loc_vs);
fs = generate_program(fs_template, glsl_version, GL_FRAGMENT_SHADER,
&loc_fs);
if (vs == 0 || fs == 0)
piglit_report_result(PIGLIT_FAIL);
glGenProgramPipelines(1, &pipe);
glBindProgramPipeline(pipe);
glUseProgramStages(pipe, GL_VERTEX_SHADER_BIT, vs);
glUseProgramStages(pipe, GL_FRAGMENT_SHADER_BIT, fs);
if (!piglit_check_gl_error(0))
piglit_report_result(PIGLIT_FAIL);
}
static bool
run_test(const struct image_op_info *op,
unsigned w, unsigned h,
bool (*check)(const struct grid_info grid,
const struct image_info img,
unsigned w, unsigned h),
const char *body)
{
const struct grid_info grid =
grid_info(GL_FRAGMENT_SHADER, GL_R32UI, W, H);
const struct image_info img = image_info_for_grid(grid);
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(img, ""),
hunk("uniform IMAGE_T img;\n"),
hunk(op->hunk),
hunk(body), NULL));
bool ret = prog &&
init_fb(grid) &&
init_image(img) &&
set_uniform_int(prog, "img", 0) &&
draw_grid(set_grid_size(grid, w, h), prog) &&
check(grid, img, w, h);
glDeleteProgram(prog);
return ret;
}
static bool
run_test(const struct image_target_info *target,
const struct image_extent size)
{
const struct grid_info grid = {
GL_FRAGMENT_SHADER_BIT,
get_image_format(GL_RGBA32F),
image_optimal_extent(size)
};
const struct image_info img = {
target, grid.format, size,
image_format_epsilon(grid.format)
};
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(img, ""),
hunk("readonly uniform IMAGE_T src_img;\n"
"writeonly uniform IMAGE_T dst_img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" imageStore(dst_img, IMAGE_ADDR(idx),"
" imageLoad(src_img, IMAGE_ADDR(idx)));\n"
" return x;\n"
"}\n"), NULL));
bool ret = prog && init_fb(grid) &&
init_image(img, 0) &&
init_image(img, 1) &&
set_uniform_int(prog, "src_img", 0) &&
set_uniform_int(prog, "dst_img", 1) &&
draw_grid(grid, prog) &&
check(img);
glDeleteProgram(prog);
return ret;
}
static bool
run_test(GLbitfield shaders)
{
const struct grid_info grid = {
shaders,
get_image_format(GL_R32UI),
{ W, H, 1, 1 }
};
const struct image_info img = image_info_for_grid(grid);
GLuint prog = generate_program(
grid,
GL_VERTEX_SHADER,
generate_source(grid, img, GL_VERTEX_SHADER),
GL_TESS_CONTROL_SHADER,
generate_source(grid, img, GL_TESS_CONTROL_SHADER),
GL_TESS_EVALUATION_SHADER,
generate_source(grid, img, GL_TESS_EVALUATION_SHADER),
GL_GEOMETRY_SHADER,
generate_source(grid, img, GL_GEOMETRY_SHADER),
GL_FRAGMENT_SHADER,
generate_source(grid, img, GL_FRAGMENT_SHADER),
GL_COMPUTE_SHADER,
generate_source(grid, img, GL_COMPUTE_SHADER));
bool ret = prog && init_fb(grid) &&
init_images(img) &&
bind_images(grid, prog) &&
draw_grid(grid, prog) &&
check(grid, img);
glDeleteProgram(prog);
return ret;
}
/**
* Copy from a source image into a destination image of the specified
* format and check the result.
*
* If \a strict_layout_qualifiers is false, uniform layout qualifiers
* will be omitted where allowed by the spec. If \a
* strict_access_qualifiers is false, the "readonly" and "writeonly"
* qualifiers will be omitted. If \a strict_binding is false, the
* image will be bound as READ_WRITE, otherwise only the required
* access type will be used.
*/
static bool
run_test(const struct image_format_info *format,
bool strict_layout_qualifiers,
bool strict_access_qualifiers,
bool strict_binding)
{
const struct grid_info grid =
grid_info(GL_FRAGMENT_SHADER,
image_base_internal_format(format), W, H);
const struct image_info img =
image_info(GL_TEXTURE_2D, format->format, W, H);
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(img, ""),
test_hunk(strict_layout_qualifiers,
strict_access_qualifiers),
hunk("SRC_IMAGE_Q uniform IMAGE_BARE_T src_img;\n"
"DST_IMAGE_Q uniform IMAGE_BARE_T dst_img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" imageStore(dst_img, IMAGE_ADDR(idx),"
" imageLoad(src_img, IMAGE_ADDR(idx)));\n"
" return x;\n"
"}\n"), NULL));
bool ret = prog && init_fb(grid) &&
init_image(img, 0, strict_binding) &&
init_image(img, 1, strict_binding) &&
set_uniform_int(prog, "src_img", 0) &&
set_uniform_int(prog, "dst_img", 1) &&
draw_grid(grid, prog) &&
check(grid, img);
glDeleteProgram(prog);
return ret;
}
static bool
run_test(const struct image_qualifier_info *qual,
const struct image_stage_info *stage_w,
const struct image_stage_info *stage_r,
unsigned l)
{
const struct grid_info grid = {
stage_w->bit | stage_r->bit,
get_image_format(GL_RGBA32UI),
{ l, l, 1, 1 }
};
const struct image_info img = image_info_for_grid(grid);
GLuint prog = generate_program(
grid,
/*
* Write (11, 22, 33, 44) to some location on the
* image from the write stage.
*/
stage_w->stage,
concat(qualifier_hunk(qual),
image_hunk(img, ""),
hunk("IMAGE_Q uniform IMAGE_T img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" imageStore(img, idx, DATA_T(11, 22, 33, 44));"
" return x;"
"}\n"), NULL),
/*
* The same location will read back the expected value
* if image access is coherent, as the shader inputs
* of the read stage are dependent on the outputs of
* the write stage and consequently they are
* guaranteed to be executed sequentially.
*/
stage_r->stage,
concat(qualifier_hunk(qual),
image_hunk(img, ""),
hunk("IMAGE_Q uniform IMAGE_T img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" DATA_T v = imageLoad(img, idx);"
" if (v == DATA_T(11, 22, 33, 44))"
" return GRID_T(33, 33, 33, 33);"
" else"
" return GRID_T(77, 77, 77, 77);"
"}\n"), NULL));
bool ret = prog &&
init_fb(grid) &&
init_image(img) &&
set_uniform_int(prog, "img", 0) &&
draw_grid(grid, prog) &&
(check(grid, img) || qual->control_test);
glDeleteProgram(prog);
return ret;
}
/*
* Compilation of the nondeterministic program.
*
* First of all, we must generate a linear system Ax = b
* containing all the constraints on rows, columns, etc.
* Then, the system is reduced to upper-triangular form
* by Gauss-Jordan elimination. This leaves some free variabled
* that the program must guess. The order of the guesses is
* chosen so to maximize the number of deductions.
*/
void compile_program(INSTR * pc)
{
/* A has 2 * n + 2 equations and nsquare columns (variables) */
int *A = alloca(nsquare * (2 * n + 2) * sizeof(int));
int *b = alloca((2 * n + 2) * sizeof(int));
int neq;
void generate_program(INSTR * pc, int *A, int *b, int r, int c);
void gauss_jordanize(int *A, int *b, int r, int c);
int generate_equations(int *A, int *b);
neq = generate_equations(A, b);
gauss_jordanize(A, b, neq, nsquare);
generate_program(pc, A, b, neq, nsquare);
}
static bool
run_test(const struct image_qualifier_info *qual)
{
const struct grid_info grid =
grid_info(GL_FRAGMENT_SHADER, GL_R32UI, W, H);
const struct image_info img =
image_info(GL_TEXTURE_1D, GL_R32UI, W, H);
GLuint prog = generate_program(
grid,
/**
* Write to consecutive locations of an image using a
* the value read from a fixed location of a different
* image uniform which aliases the first image. If
* the implementation incorrectly coalesces repeated
* loads from the fixed location the results of the
* test will be altered.
*/
GL_FRAGMENT_SHADER,
concat(qualifier_hunk(qual),
image_hunk(img, ""),
hunk("IMAGE_Q IMAGE_UNIFORM_T src_img;\n"
"IMAGE_Q IMAGE_UNIFORM_T dst_img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" int i;\n"
"\n"
" for (i = 0; i < N / 2; ++i) {\n"
" imageStore(dst_img, 2 * i,"
" imageLoad(src_img, W) + 1u);\n"
" imageStore(dst_img, 2 * i + 1,"
" imageLoad(src_img, W) - 1u);\n"
" }\n"
"\n"
" return x;\n"
"}\n"), NULL));
bool ret = prog &&
init_fb(grid) &&
init_image(img) &&
set_uniform_int(prog, "src_img", 0) &&
set_uniform_int(prog, "dst_img", 0) &&
draw_grid(set_grid_size(grid, 1, 1), prog) &&
(check(img) || qual->control_test);
glDeleteProgram(prog);
return ret;
}
int
main ( int argc, char **argv )
{
yyparse();
simplify_tree ( &root, root );
// node_print(root, 0);
find_globals();
size_t n_globals = tlhash_size(global_names);
symbol_t *global_list[n_globals];
tlhash_values ( global_names, (void **)&global_list );
for ( size_t i=0; i<n_globals; i++ )
if ( global_list[i]->type == SYM_FUNCTION )
bind_names ( global_list[i], global_list[i]->node );
// print_globals();
generate_program ();
destroy_subtree ( root );
destroy_symtab();
}
/**
* Test skeleton: Init image to \a init_value, run the provided shader
* \a op and check that the resulting image pixels equal \a
* check_value.
*/
static bool
run_test(uint32_t init_value, uint32_t check_value,
const char *op)
{
const struct grid_info grid =
grid_info(GL_FRAGMENT_SHADER, GL_R32UI, W, H);
const struct image_info img = image_info_for_grid(grid);
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(img, ""),
hunk("uniform IMAGE_T img;\n"),
hunk(op), NULL));
bool ret = prog &&
init_fb(grid) &&
init_image(img, init_value) &&
set_uniform_int(prog, "img", 0) &&
draw_grid(grid, prog) &&
check(img, check_value);
glDeleteProgram(prog);
return ret;
}
/**
* If \a layered is false, bind an individual layer of a texture to an
* image unit, read its contents and write back a different value to
* the same location. If \a layered is true or the texture has a
* single layer, the whole texture will be read and written back.
*
* For textures with a single layer, the arguments \a layered and \a
* layer which are passed to the same arguments of
* glBindImageTexture() should have no effect as required by the spec.
*/
static bool
run_test(const struct image_target_info *target,
bool layered, unsigned layer)
{
const struct image_info real_img = image_info(
target->target, GL_RGBA32F, W, H);
const unsigned slices = (layered ? 1 : image_num_layers(real_img));
/*
* "Slice" of the image that will be bound to the pipeline.
*/
const struct image_info slice_img = image_info(
(layered ? target->target : image_layer_target(target)),
GL_RGBA32F, W, H / slices);
/*
* Grid with as many elements as the slice.
*/
const struct grid_info grid = grid_info(
GL_FRAGMENT_SHADER, GL_RGBA32F, W, H / slices);
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(slice_img, ""),
hunk("IMAGE_UNIFORM_T img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" GRID_T v = imageLoad(img, IMAGE_ADDR(idx));\n"
" imageStore(img, IMAGE_ADDR(idx), DATA_T(33));\n"
" return v;\n"
"}\n"), NULL));
bool ret = prog && init_fb(grid) &&
init_image(real_img, layered, layer) &&
set_uniform_int(prog, "img", 0) &&
draw_grid(grid, prog) &&
check(grid, real_img, (slices == 1 ? 0 : layer));
glDeleteProgram(prog);
return ret;
}
int collisionfinding(parameters_type& parameters)
{
bool usetunnelbitconditions = parameters.usetunnelbitconditions;
sha1messagespace tmpspace;
vector< vector<uint32> > bitrels, tmpbitrel, tmpbitrel2;
for (unsigned i = 0; i < parameters.rnd234_m_bitrelationfiles.size(); ++i) {
try {
cout << "Loading '" << parameters.rnd234_m_bitrelationfiles[i] << "'..." << flush;
load_bz2(tmpspace, text_archive, parameters.rnd234_m_bitrelationfiles[i]);
cout << "done" << flush;
tmpspace.tobitrelations_80(tmpbitrel);
bitrels.insert(bitrels.end(), tmpbitrel.begin(), tmpbitrel.end());
cout << " (" << tmpbitrel.size() << " bitrels, new total: " << bitrels.size() << ")" << endl;
} catch (std::exception& e) {
cerr << "Exception:" << endl << e.what() << endl;
return 1;
}
}
vector< sha1differentialpath > paths;
cout << "Loading round 1 paths from '" << parameters.rnd1_pathsfile << "'..." << flush;
try {
load_bz2(paths, text_archive, parameters.rnd1_pathsfile);
cout << "done (" << paths.size() << " paths)." << endl;
} catch (std::exception& e) {
cerr << "Exception:" << endl << e.what() << endl;
return 1;
}
if (paths.size() == 0) exit(0);
vector< vector<uint32> > bitrelsbu = bitrels;
for (unsigned pi = 0; pi < paths.size(); ++pi) { try {
bitrels = bitrelsbu;
sha1differentialpath upperpath;
cout << "Loading round 2,3,4 path from '" << parameters.rnd234_pathfile << "'..." << flush;
try {
load_bz2(upperpath, text_archive, parameters.rnd234_pathfile);
cout << "done." << endl;
} catch (std::exception& e) {
cerr << "Exception:" << endl << e.what() << endl;
return 1;
}
for (int t = paths[pi].tbegin(); t < paths[pi].tend(); ++t)
upperpath[t] = paths[pi][t];
for (unsigned t = 0; t < paths[pi].tend()-1; ++t)
upperpath.getme(t) = paths[pi].getme(t);
cleanup_path(upperpath);
maindiffpath = upperpath;
bool bad = false;
for (int t = maindiffpath.tbegin()+4; t < maindiffpath.tend()-1; ++t) {
if (maindiffpath.getme(t).mask != parameters.m_mask[t]) {
uint32 dm = maindiffpath.getme(t).adddiff();
uint32 mcur = 0;
uint32 mmask = parameters.m_mask[t];
uint32 madd = (~mmask)+1;
sdr msdr;
msdr.mask = mmask;
do {
mcur += madd; mcur &= mmask;
msdr.sign = mcur;
if (msdr.adddiff() == dm) {
cout << "corrected: \t" << t << ":\t" << msdr << " == " << sdr(parameters.m_mask[t]) << endl;
break;
}
} while (mcur != 0);
if (msdr.adddiff() == dm) {
maindiffpath.getme(t) = msdr;
} else {
bad = true;
cout << "failed: \t" << t << ":\t" << maindiffpath.getme(t) << " != " << sdr(parameters.m_mask[t]) << endl;
}
}
}
if (bad) exit(0);
show_path(maindiffpath);
// remove bitrelations possibly limiting me[0],...,me[19]
const unsigned tend_fix_me = parameters.tend_rnd1_me;
cout << "Fixed me diffs for t=[0," << tend_fix_me << "): (" << bitrels.size() << "=>" << flush;
filter_bitconditions(bitrels, tend_fix_me, 80);
cout << bitrels.size() << ")" << endl;
for (unsigned i = 0; i < bitrels.size(); ++i) {
cout << " - ";
bool firstone = true;
for (unsigned t = 0; t < 80; ++t)
for (unsigned b = 0; b < 32; ++b)
if (bitrels[i][t] & (1<<b)) {
if (firstone)
firstone = false;
else
cout << " + ";
cout << "M[" << t << "," << b << "]";
}
cout << " = " << (bitrels[i][80]&1) << endl;
}
tmpspace.clear();
for (unsigned t = 0; t < 80; ++t) {
if (t < tend_fix_me) {
for (unsigned b = 0; b < 31; ++b) {
int bit = maindiffpath.getme(t).get(b);
if (bit == 0)
tmpspace.buildbasis_addfreebit(t,b);
else
tmpspace.buildbasis_setbit(t,b,bit==-1);
}
tmpspace.buildbasis_addfreebit(t,31);
} else {
for (unsigned b = 0; b < 32; ++b)
tmpspace.buildbasis_addfreebit(t,b);
}
}
tmpspace.tobitrelations_80(tmpbitrel);
bitrels.insert(bitrels.end(), tmpbitrel.begin(), tmpbitrel.end());
cout << "Extra me [0," << tend_fix_me << ") bitrelations: " << tmpbitrel.size() << endl;
{
cout << "Extra tunnel me bitrelations: " << endl;
ifstream ifs("tunnel_bitconditions.txt");
read_message_bitconditions(ifs, tmpbitrel);
for (unsigned i = 0; i < tmpbitrel.size(); ++i) {
bool firstone = true;
for (unsigned t = 0; t < 80; ++t)
if (tmpbitrel[i][t]) {
if (firstone) firstone = false;
else cout << "+ ";
cout << "m" << t << sdr(tmpbitrel[i][t]) << " ";
}
cout << "= " << (tmpbitrel[i][80]&1) << endl;
}
bitrels.insert(bitrels.end(), tmpbitrel.begin(), tmpbitrel.end());
}
tmpspace.frombitrelations_80(bitrels);
mainmespace = tmpspace;
tmpspace.tobitrelations_16(bitrels);
cout << "Total bitrelations: " << bitrels.size() << endl;
#if 0
cout << "Independent message bits: " << endl;
uint32 meindep[16];
for (unsigned i = 0; i < 16; ++i)
meindep[16] = ~uint32(0);
for (unsigned i = 0; i < bitrels.size(); ++i)
for (unsigned t = 0; t < 16; ++t)
meindep[t] &= ~bitrels[i][t];
for (unsigned i = 0; i < 16; ++i) {
cout << "m[" << i << "] indep:\t";
for (int b = 31; b >= 0; --b)
cout << ((meindep[i]&(1<<b))?"1":"0");
cout << endl;
}
#endif
pathbitrelations = bitrels;
pathbitrelationsmatrix.clear();
pathbitrelationsmatrix.resize(16);
for (unsigned i = 0; i < 16; ++i)
pathbitrelationsmatrix[i].resize(32);
for (unsigned i = 0; i < pathbitrelations.size(); ++i) {
int lastcol = -1;
for (int col = 16*32-1; col >= 0; --col)
if (pathbitrelations[i][col>>5]&(1<<(col&31))) {
lastcol = col;
unsigned t = lastcol>>5;
unsigned b = lastcol&31;
pathbitrelationsmatrix[t][b] = pathbitrelations[i];
break;
}
if (lastcol == -1) throw;
}
#if 0
//needs diffpath & pathbitrelationsmatrix
analyze_indepsection_prob();
#endif
if (parameters.tunnelfile.size()) {
vector<sha1differentialpath> tunnels;
cout << "Loading tunnels from '" << parameters.tunnelfile << "'..." << flush;
try {
load_bz2(tunnels, text_archive, parameters.tunnelfile);
}
catch (std::exception &e) { tunnels.clear(); cout << "failed:" << endl << e.what() << endl; }
catch (...) { tunnels.clear(); cout << "failed." << endl; }
if (tunnels.size())
analyze_tunnels_diffpath(maindiffpath, bitrels, tunnels);
exit(0);
}
if (!usetunnelbitconditions) {
// if not using tunnel bit conditions then we'll assume we want to analyze tunnels
analyze_tunnels_diffpath(maindiffpath, bitrels);
exit(0);
}
// if we're using tunnel bit conditions we'll assume we want to generate the collision finding program
generate_program();
break;
} catch (std::exception& e) { cerr << "c: " << e.what() << endl; } catch (...) {}
} // for (unsigned pi = 0; pi < paths.size(); ++pi)
}
/**
* Test binding image uniforms to image units for a simple shader
* program.
*/
static bool
run_test_uniform(void)
{
const struct grid_info grid =
grid_info(GL_FRAGMENT_SHADER, GL_RGBA32F, W, H);
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(image_info_for_grid(grid), ""),
hunk("uniform IMAGE_T imgs[2];\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" imageStore(imgs[0], IMAGE_ADDR(idx), x);\n"
" imageStore(imgs[1], IMAGE_ADDR(idx), x);\n"
" return x;\n"
"}\n"), NULL));
const int loc = glGetUniformLocation(prog, "imgs");
bool ret = prog && check_uniform_int(prog, loc, 0) &&
check_uniform_int(prog, loc + 1, 0);
int v[2];
glUseProgram(prog);
/*
* Image uniforms are bound to image units using
* glUniform1i{v}.
*/
glUniform1i(loc, 3);
ret &= check_uniform_int(prog, loc, 3) &&
check_uniform_int(prog, loc + 1, 0);
glUniform1i(loc + 1, 3);
ret &= check_uniform_int(prog, loc, 3) &&
check_uniform_int(prog, loc + 1, 3);
v[0] = 4;
v[1] = 5;
glUniform1iv(loc, 2, v);
ret &= check_uniform_int(prog, loc, 4) &&
check_uniform_int(prog, loc + 1, 5);
/*
* GL_INVALID_VALUE is generated if the value specified is
* greater than or equal to the value of GL_MAX_IMAGE_UNITS.
*/
glUniform1i(loc, max_image_units());
ret &= piglit_check_gl_error(GL_INVALID_VALUE);
v[0] = 3;
v[1] = max_image_units() + 1;
glUniform1iv(loc, 2, v);
ret &= piglit_check_gl_error(GL_INVALID_VALUE);
/*
* GL_INVALID_VALUE is generated if the value specified is
* less than zero.
*/
glUniform1i(loc, -1);
ret &= piglit_check_gl_error(GL_INVALID_VALUE);
v[0] = 3;
v[1] = -4;
glUniform1iv(loc, 2, v);
ret &= piglit_check_gl_error(GL_INVALID_VALUE);
/*
* GL_INVALID_OPERATION is generated by Uniform* functions
* other than Uniform1i{v}.
*/
CHECK_INVAL_2(glUniform, 1f, 1ui, (loc, 0), ret);
CHECK_INVAL_3(glUniform, 2i, 2f, 2ui, (loc, 0, 0), ret);
CHECK_INVAL_3(glUniform, 3i, 3f, 3ui, (loc, 0, 0, 0), ret);
CHECK_INVAL_3(glUniform, 4i, 4f, 4ui, (loc, 0, 0, 0, 0), ret);
CHECK_INVAL_2(glUniform, 1fv, 1uiv, (loc, 1, (void *)v), ret);
CHECK_INVAL_3(glUniform, 2iv, 2fv, 2uiv, (loc, 1, (void *)v), ret);
CHECK_INVAL_3(glUniform, 3iv, 3fv, 3uiv, (loc, 1, (void *)v), ret);
CHECK_INVAL_3(glUniform, 4iv, 4fv, 4uiv, (loc, 1, (void *)v), ret);
CHECK_INVAL_3(glUniformMatrix, 2fv, 3fv, 4fv,
(loc, 1, GL_FALSE, (float *)v), ret);
CHECK_INVAL_3(glUniformMatrix, 2x3fv, 3x2fv, 2x4fv,
(loc, 1, GL_FALSE, (float *)v), ret);
CHECK_INVAL_3(glUniformMatrix, 4x2fv, 3x4fv, 4x3fv,
(loc, 1, GL_FALSE, (float *)v), ret);
if (piglit_is_extension_supported("GL_ARB_gpu_shader_fp64")) {
CHECK_INVAL_1(glUniform, 1d, (loc, 0), ret);
CHECK_INVAL_1(glUniform, 2d, (loc, 0, 0), ret);
CHECK_INVAL_1(glUniform, 3d, (loc, 0, 0, 0), ret);
CHECK_INVAL_1(glUniform, 4d, (loc, 0, 0, 0, 0), ret);
CHECK_INVAL_2(glUniform, 1dv, 2dv, (loc, 1, (double *)v), ret);
CHECK_INVAL_2(glUniform, 3dv, 4dv, (loc, 1, (double *)v), ret);
CHECK_INVAL_3(glUniformMatrix, 2dv, 3dv, 4dv,
(loc, 1, GL_FALSE, (double *)v), ret);
CHECK_INVAL_3(glUniformMatrix, 2x3dv, 3x2dv, 2x4dv,
(loc, 1, GL_FALSE, (double *)v), ret);
CHECK_INVAL_3(glUniformMatrix, 4x2dv, 3x4dv, 4x3dv,
(loc, 1, GL_FALSE, (double *)v), ret);
}
glDeleteProgram(prog);
return ret;
}
bool
download_image_levels(const struct image_info img, unsigned num_levels,
unsigned unit, uint32_t *r_pixels)
{
const unsigned m = image_num_components(img.format);
int i, l;
glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT |
GL_BUFFER_UPDATE_BARRIER_BIT |
GL_PIXEL_BUFFER_BARRIER_BIT |
GL_SHADER_IMAGE_ACCESS_BARRIER_BIT);
glBindTexture(img.target->target, textures[unit]);
switch (img.target->target) {
case GL_TEXTURE_1D:
case GL_TEXTURE_2D:
case GL_TEXTURE_3D:
case GL_TEXTURE_RECTANGLE:
case GL_TEXTURE_1D_ARRAY:
case GL_TEXTURE_2D_ARRAY:
case GL_TEXTURE_CUBE_MAP_ARRAY:
assert(img.target->target != GL_TEXTURE_RECTANGLE ||
num_levels == 1);
for (l = 0; l < num_levels; ++l)
glGetTexImage(img.target->target, l,
img.format->pixel_format,
image_base_type(img.format),
&r_pixels[m * image_level_offset(img, l)]);
break;
case GL_TEXTURE_CUBE_MAP:
for (l = 0; l < num_levels; ++l) {
const unsigned offset = m * image_level_offset(img, l);
const unsigned face_sz =
m * product(image_level_size(img, l)) / 6;
for (i = 0; i < 6; ++i)
glGetTexImage(GL_TEXTURE_CUBE_MAP_POSITIVE_X + i,
l, img.format->pixel_format,
image_base_type(img.format),
&r_pixels[offset + face_sz * i]);
}
break;
case GL_TEXTURE_BUFFER: {
/*
* glGetTexImage() isn't supposed to work with buffer
* textures. We copy the packed pixels to a texture
* with the same internal format as the image to let
* the GL unpack it for us.
*/
const struct image_extent grid = image_optimal_extent(img.size);
GLuint packed_tex;
assert(num_levels == 1);
glGenTextures(1, &packed_tex);
glBindTexture(GL_TEXTURE_2D, packed_tex);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, buffers[unit]);
glTexImage2D(GL_TEXTURE_2D, 0, img.format->format,
grid.x, grid.y, 0, img.format->pixel_format,
img.format->pixel_type, NULL);
glGetTexImage(GL_TEXTURE_2D, 0, img.format->pixel_format,
image_base_type(img.format), r_pixels);
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
glDeleteTextures(1, &packed_tex);
break;
}
case GL_TEXTURE_2D_MULTISAMPLE:
case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: {
/*
* GL doesn't seem to provide any direct way to read
* back a multisample texture, so we use imageLoad()
* to copy its contents to a larger single-sample 2D
* texture from the fragment shader.
*/
const struct grid_info grid = {
get_image_stage(GL_FRAGMENT_SHADER)->bit,
img.format,
image_optimal_extent(img.size)
};
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(img, "SRC_"),
image_hunk(image_info_for_grid(grid), "DST_"),
hunk("readonly SRC_IMAGE_UNIFORM_T src_img;\n"
"writeonly DST_IMAGE_UNIFORM_T dst_img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" imageStore(dst_img, DST_IMAGE_ADDR(idx),\n"
" imageLoad(src_img, SRC_IMAGE_ADDR(idx)));\n"
" return x;\n"
"}\n"), NULL));
bool ret = prog && generate_fb(grid, 1);
GLuint tmp_tex;
assert(num_levels == 1);
glGenTextures(1, &tmp_tex);
glBindTexture(GL_TEXTURE_2D, tmp_tex);
glTexImage2D(GL_TEXTURE_2D, 0, img.format->format,
grid.size.x, grid.size.y, 0,
img.format->pixel_format, image_base_type(img.format),
NULL);
glBindImageTexture(unit, textures[unit], 0, GL_TRUE, 0,
GL_READ_ONLY, img.format->format);
glBindImageTexture(6, tmp_tex, 0, GL_TRUE, 0,
GL_WRITE_ONLY, img.format->format);
ret &= set_uniform_int(prog, "src_img", unit) &&
set_uniform_int(prog, "dst_img", 6) &&
draw_grid(grid, prog);
glMemoryBarrier(GL_TEXTURE_UPDATE_BARRIER_BIT);
glGetTexImage(GL_TEXTURE_2D, 0, img.format->pixel_format,
image_base_type(img.format), r_pixels);
glDeleteProgram(prog);
glDeleteTextures(1, &tmp_tex);
glBindFramebuffer(GL_FRAMEBUFFER, fb[0]);
glViewportIndexedfv(0, vp[0]);
if (!ret)
return false;
break;
}
default:
abort();
}
return piglit_check_gl_error(GL_NO_ERROR);
}
bool
upload_image_levels(const struct image_info img, unsigned num_levels,
unsigned level, unsigned unit, const uint32_t *pixels)
{
const unsigned m = image_num_components(img.format);
int i, l;
if (get_texture(unit)) {
glDeleteTextures(1, &textures[unit]);
textures[unit] = 0;
}
if (get_buffer(unit)) {
glDeleteBuffers(1, &buffers[unit]);
buffers[unit] = 0;
}
glGenTextures(1, &textures[unit]);
glBindTexture(img.target->target, textures[unit]);
switch (img.target->target) {
case GL_TEXTURE_1D:
for (l = 0; l < num_levels; ++l) {
const struct image_extent size = image_level_size(img, l);
glTexImage1D(GL_TEXTURE_1D, l, img.format->format,
size.x, 0, img.format->pixel_format,
image_base_type(img.format),
&pixels[m * image_level_offset(img, l)]);
}
break;
case GL_TEXTURE_2D:
for (l = 0; l < num_levels; ++l) {
const struct image_extent size = image_level_size(img, l);
glTexImage2D(GL_TEXTURE_2D, l, img.format->format,
size.x, size.y, 0,
img.format->pixel_format,
image_base_type(img.format),
&pixels[m * image_level_offset(img, l)]);
}
break;
case GL_TEXTURE_3D:
for (l = 0; l < num_levels; ++l) {
const struct image_extent size = image_level_size(img, l);
glTexImage3D(GL_TEXTURE_3D, l, img.format->format,
size.x, size.y, size.z, 0,
img.format->pixel_format,
image_base_type(img.format),
&pixels[m * image_level_offset(img, l)]);
}
break;
case GL_TEXTURE_RECTANGLE:
assert(num_levels == 1);
glTexImage2D(GL_TEXTURE_RECTANGLE, 0, img.format->format,
img.size.x, img.size.y, 0, img.format->pixel_format,
image_base_type(img.format), pixels);
break;
case GL_TEXTURE_CUBE_MAP:
for (l = 0; l < num_levels; ++l) {
const unsigned offset = m * image_level_offset(img, l);
const struct image_extent size = image_level_size(img, l);
const unsigned face_sz = m * product(size) / 6;
for (i = 0; i < 6; ++i)
glTexImage2D(GL_TEXTURE_CUBE_MAP_POSITIVE_X + i, l,
img.format->format, size.x, size.y, 0,
img.format->pixel_format,
image_base_type(img.format),
&pixels[offset + face_sz * i]);
}
break;
case GL_TEXTURE_BUFFER: {
/*
* glTexImage*() isn't supposed to work with buffer
* textures. We copy the unpacked pixels to a texture
* with the desired internal format to let the GL pack
* them for us.
*/
const struct image_extent grid = image_optimal_extent(img.size);
GLuint packed_tex;
assert(num_levels == 1);
glGenBuffers(1, &buffers[unit]);
glBindBuffer(GL_PIXEL_PACK_BUFFER, buffers[unit]);
glBufferData(GL_PIXEL_PACK_BUFFER,
img.size.x * image_pixel_size(img.format) / 8,
NULL, GL_STATIC_DRAW);
glGenTextures(1, &packed_tex);
glBindTexture(GL_TEXTURE_2D, packed_tex);
glTexImage2D(GL_TEXTURE_2D, 0, img.format->format,
grid.x, grid.y, 0, img.format->pixel_format,
image_base_type(img.format), pixels);
glGetTexImage(GL_TEXTURE_2D, 0, img.format->pixel_format,
img.format->pixel_type, NULL);
glDeleteTextures(1, &packed_tex);
glBindBuffer(GL_PIXEL_PACK_BUFFER, 0);
glTexBuffer(GL_TEXTURE_BUFFER, image_compat_format(img.format),
buffers[unit]);
break;
}
case GL_TEXTURE_1D_ARRAY:
for (l = 0; l < num_levels; ++l) {
const struct image_extent size = image_level_size(img, l);
glTexImage2D(GL_TEXTURE_1D_ARRAY, l, img.format->format,
size.x, size.y, 0, img.format->pixel_format,
image_base_type(img.format),
&pixels[m * image_level_offset(img, l)]);
}
break;
case GL_TEXTURE_2D_ARRAY:
for (l = 0; l < num_levels; ++l) {
const struct image_extent size = image_level_size(img, l);
glTexImage3D(GL_TEXTURE_2D_ARRAY, l, img.format->format,
size.x, size.y, size.z, 0,
img.format->pixel_format,
image_base_type(img.format),
&pixels[m * image_level_offset(img, l)]);
}
break;
case GL_TEXTURE_CUBE_MAP_ARRAY:
for (l = 0; l < num_levels; ++l) {
const struct image_extent size = image_level_size(img, l);
glTexImage3D(GL_TEXTURE_CUBE_MAP_ARRAY, l, img.format->format,
size.x, size.y, size.z, 0,
img.format->pixel_format,
image_base_type(img.format),
&pixels[m * image_level_offset(img, l)]);
}
break;
case GL_TEXTURE_2D_MULTISAMPLE:
case GL_TEXTURE_2D_MULTISAMPLE_ARRAY: {
/*
* GL doesn't seem to provide any direct way to
* initialize a multisample texture, so we use
* imageStore() to render to it from the fragment
* shader copying the contents of a larger
* single-sample 2D texture.
*/
const struct grid_info grid = {
get_image_stage(GL_FRAGMENT_SHADER)->bit,
img.format,
image_optimal_extent(img.size)
};
GLuint prog = generate_program(
grid, GL_FRAGMENT_SHADER,
concat(image_hunk(image_info_for_grid(grid), "SRC_"),
image_hunk(img, "DST_"),
hunk("readonly SRC_IMAGE_UNIFORM_T src_img;\n"
"writeonly DST_IMAGE_UNIFORM_T dst_img;\n"
"\n"
"GRID_T op(ivec2 idx, GRID_T x) {\n"
" imageStore(dst_img, DST_IMAGE_ADDR(idx),\n"
" imageLoad(src_img, SRC_IMAGE_ADDR(idx)));\n"
" return x;\n"
"}\n"), NULL));
bool ret = prog && generate_fb(grid, 1);
GLuint tmp_tex;
assert(num_levels == 1);
glGenTextures(1, &tmp_tex);
glBindTexture(GL_TEXTURE_2D, tmp_tex);
if (img.target->target == GL_TEXTURE_2D_MULTISAMPLE_ARRAY) {
glTexImage3DMultisample(GL_TEXTURE_2D_MULTISAMPLE_ARRAY,
img.size.x, img.format->format,
img.size.y, img.size.z, img.size.w,
GL_FALSE);
} else {
glTexImage2DMultisample(GL_TEXTURE_2D_MULTISAMPLE,
img.size.x, img.format->format,
img.size.y, img.size.z,
GL_FALSE);
}
glTexImage2D(GL_TEXTURE_2D, 0, img.format->format,
grid.size.x, grid.size.y, 0,
img.format->pixel_format, image_base_type(img.format),
pixels);
glBindImageTexture(unit, textures[unit], 0, GL_TRUE, 0,
GL_WRITE_ONLY, img.format->format);
glBindImageTexture(6, tmp_tex, 0, GL_TRUE, 0,
GL_READ_ONLY, img.format->format);
ret &= set_uniform_int(prog, "src_img", 6) &&
set_uniform_int(prog, "dst_img", unit) &&
draw_grid(grid, prog);
glDeleteProgram(prog);
glDeleteTextures(1, &tmp_tex);
glBindFramebuffer(GL_FRAMEBUFFER, fb[0]);
glViewportIndexedfv(0, vp[0]);
glMemoryBarrier(GL_SHADER_IMAGE_ACCESS_BARRIER_BIT);
if (!ret)
return false;
break;
}
default:
abort();
}
glBindImageTexture(unit, textures[unit], level, GL_TRUE, 0,
GL_READ_WRITE, img.format->format);
return piglit_check_gl_error(GL_NO_ERROR);
}
