static bool
brw_codegen_tcs_prog(struct brw_context *brw,
struct gl_shader_program *shader_prog,
struct brw_program *tcp,
struct brw_tcs_prog_key *key)
{
struct gl_context *ctx = &brw->ctx;
const struct brw_compiler *compiler = brw->screen->compiler;
const struct gen_device_info *devinfo = compiler->devinfo;
struct brw_stage_state *stage_state = &brw->tcs.base;
nir_shader *nir;
struct brw_tcs_prog_data prog_data;
bool start_busy = false;
double start_time = 0;
void *mem_ctx = ralloc_context(NULL);
if (tcp) {
nir = tcp->program.nir;
} else {
/* Create a dummy nir_shader. We won't actually use NIR code to
* generate assembly (it's easier to generate assembly directly),
* but the whole compiler assumes one of these exists.
*/
const nir_shader_compiler_options *options =
ctx->Const.ShaderCompilerOptions[MESA_SHADER_TESS_CTRL].NirOptions;
nir = create_passthrough_tcs(mem_ctx, compiler, options, key);
}
memset(&prog_data, 0, sizeof(prog_data));
/* Allocate the references to the uniforms that will end up in the
* prog_data associated with the compiled program, and which will be freed
* by the state cache.
*
* Note: param_count needs to be num_uniform_components * 4, since we add
* padding around uniform values below vec4 size, so the worst case is that
* every uniform is a float which gets padded to the size of a vec4.
*/
int param_count = nir->num_uniforms / 4;
prog_data.base.base.param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.base.pull_param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.base.nr_params = param_count;
if (tcp) {
brw_assign_common_binding_table_offsets(MESA_SHADER_TESS_CTRL, devinfo,
shader_prog, &tcp->program,
&prog_data.base.base, 0);
prog_data.base.base.image_param =
rzalloc_array(NULL, struct brw_image_param,
tcp->program.info.num_images);
prog_data.base.base.nr_image_params = tcp->program.info.num_images;
brw_nir_setup_glsl_uniforms(nir, shader_prog, &tcp->program,
&prog_data.base.base,
compiler->scalar_stage[MESA_SHADER_TESS_CTRL]);
} else {
TEST(ir_variable_constructor, interface_array)
{
void *mem_ctx = ralloc_context(NULL);
static const glsl_struct_field f[] = {
{
glsl_type::vec(4),
"v",
false
}
};
const glsl_type *const interface =
glsl_type::get_interface_instance(f,
ARRAY_SIZE(f),
GLSL_INTERFACE_PACKING_STD140,
"simple_interface");
const glsl_type *const interface_array =
glsl_type::get_array_instance(interface, 2);
static const char name[] = "array_instance";
ir_variable *const v =
new(mem_ctx) ir_variable(interface_array, name, ir_var_uniform);
EXPECT_STREQ(name, v->name);
EXPECT_NE(name, v->name);
EXPECT_EQ(interface_array, v->type);
EXPECT_EQ(interface, v->get_interface_type());
}
brw_blorp_eu_emitter::brw_blorp_eu_emitter(struct brw_context *brw,
bool debug_flag)
: mem_ctx(ralloc_context(NULL)),
generator(brw->intelScreen->compiler, brw,
mem_ctx, (void *) rzalloc(mem_ctx, struct brw_wm_prog_key),
(struct brw_stage_prog_data *) rzalloc(mem_ctx, struct brw_wm_prog_data),
0, false, MESA_SHADER_FRAGMENT)
{
if (debug_flag)
generator.enable_debug("blorp");
}
brw_blorp_eu_emitter::~brw_blorp_eu_emitter()
{
ralloc_free(mem_ctx);
}
const unsigned *
brw_blorp_eu_emitter::get_program(unsigned *program_size)
{
cfg_t cfg(&insts);
generator.generate_code(&cfg, 16);
return generator.get_assembly(program_size);
}
output_read_remover::output_read_remover(unsigned stage)
{
this->stage = stage;
mem_ctx = ralloc_context(NULL);
replacements = _mesa_hash_table_create(NULL, hash_table_var_hash,
_mesa_key_pointer_equal);
}
loop_state::loop_state()
{
this->ht = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
this->mem_ctx = ralloc_context(NULL);
this->loop_found = false;
}
loop_state::loop_state()
{
this->ht = hash_table_ctor(0, hash_table_pointer_hash,
hash_table_pointer_compare);
this->mem_ctx = ralloc_context(NULL);
this->loop_found = false;
}
has_recursion_visitor()
: current(NULL)
{
this->mem_ctx = ralloc_context(NULL);
this->function_hash = hash_table_ctor(0, hash_table_pointer_hash,
hash_table_pointer_compare);
}
static bool
brw_codegen_cs_prog(struct brw_context *brw,
struct gl_shader_program *prog,
struct brw_compute_program *cp,
struct brw_cs_prog_key *key)
{
struct gl_context *ctx = &brw->ctx;
const GLuint *program;
void *mem_ctx = ralloc_context(NULL);
GLuint program_size;
struct brw_cs_prog_data prog_data;
struct gl_shader *cs = prog->_LinkedShaders[MESA_SHADER_COMPUTE];
assert (cs);
memset(&prog_data, 0, sizeof(prog_data));
/* Allocate the references to the uniforms that will end up in the
* prog_data associated with the compiled program, and which will be freed
* by the state cache.
*/
int param_count = cs->num_uniform_components +
cs->NumImages * BRW_IMAGE_PARAM_SIZE;
/* The backend also sometimes adds params for texture size. */
param_count += 2 * ctx->Const.Program[MESA_SHADER_COMPUTE].MaxTextureImageUnits;
prog_data.base.param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.pull_param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
prog_data.base.image_param =
rzalloc_array(NULL, struct brw_image_param, cs->NumImages);
prog_data.base.nr_params = param_count;
prog_data.base.nr_image_params = cs->NumImages;
program = brw_cs_emit(brw, mem_ctx, key, &prog_data,
&cp->program, prog, &program_size);
if (program == NULL) {
ralloc_free(mem_ctx);
return false;
}
if (prog_data.base.total_scratch) {
brw_get_scratch_bo(brw, &brw->cs.base.scratch_bo,
prog_data.base.total_scratch * brw->max_cs_threads);
}
if (unlikely(INTEL_DEBUG & DEBUG_CS))
fprintf(stderr, "\n");
brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
key, sizeof(*key),
program, program_size,
&prog_data, sizeof(prog_data),
&brw->cs.base.prog_offset, &brw->cs.prog_data);
ralloc_free(mem_ctx);
return true;
}
has_recursion_visitor()
: current(NULL)
{
progress = false;
this->mem_ctx = ralloc_context(NULL);
this->function_hash = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
}
brw_blorp_clear_program::brw_blorp_clear_program(
struct brw_context *brw,
const brw_blorp_clear_prog_key *key)
: mem_ctx(ralloc_context(NULL)),
brw(brw),
key(key)
{
brw_init_compile(brw, &func, mem_ctx);
}
/* Test that data values are written and read with proper alignment. */
static void
test_alignment(void)
{
void *ctx = ralloc_context(NULL);
struct blob *blob;
struct blob_reader reader;
uint8_t bytes[] = "ABCDEFGHIJKLMNOP";
size_t delta, last, num_bytes;
blob = blob_create(ctx);
/* First, write an intptr value to the blob and capture that size. This is
* the expected offset between any pair of intptr values (if written with
* alignment).
*/
blob_write_intptr(blob, (intptr_t) blob);
delta = blob->size;
last = blob->size;
/* Then loop doing the following:
*
* 1. Write an unaligned number of bytes
* 2. Verify that write results in an unaligned size
* 3. Write an intptr_t value
* 2. Verify that that write results in an aligned size
*/
for (num_bytes = 1; num_bytes < sizeof(intptr_t); num_bytes++) {
blob_write_bytes(blob, bytes, num_bytes);
expect_unequal(delta, blob->size - last, "unaligned write of bytes");
blob_write_intptr(blob, (intptr_t) blob);
expect_equal(2 * delta, blob->size - last, "aligned write of intptr");
last = blob->size;
}
/* Finally, test that reading also does proper alignment. Since we know
* that values were written with all the right alignment, all we have to do
* here is verify that correct values are read.
*/
blob_reader_init(&reader, blob->data, blob->size);
expect_equal((intptr_t) blob, blob_read_intptr(&reader),
"read of initial, aligned intptr_t");
for (num_bytes = 1; num_bytes < sizeof(intptr_t); num_bytes++) {
expect_equal_bytes(bytes, blob_read_bytes(&reader, num_bytes),
num_bytes, "unaligned read of bytes");
expect_equal((intptr_t) blob, blob_read_intptr(&reader),
"aligned read of intptr_t");
}
ralloc_free(ctx);
}
get_sampler_name(ir_dereference *last,
struct gl_shader_program *shader_program)
{
this->mem_ctx = ralloc_context(NULL);
this->shader_program = shader_program;
this->name = NULL;
this->offset = 0;
this->last = last;
}
void
set_uniform_initializer::SetUp()
{
this->mem_ctx = ralloc_context(NULL);
this->prog = rzalloc(NULL, struct gl_shader_program);
/* Set default values used by the test cases.
*/
this->actual_index = 1;
this->name = "i";
}
void
link_varyings::SetUp()
{
this->mem_ctx = ralloc_context(NULL);
this->ir.make_empty();
this->consumer_inputs
= hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
this->consumer_interface_inputs
= hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
}
brw_blorp_const_color_program::brw_blorp_const_color_program(
struct brw_context *brw,
const brw_blorp_const_color_prog_key *key)
: mem_ctx(ralloc_context(NULL)),
brw(brw),
key(key),
R0(),
R1(),
clear_rgba(),
base_mrf(0)
{
brw_init_compile(brw, &func, mem_ctx);
}
static void
brw_vs_init_compile(struct brw_context *brw,
struct gl_shader_program *prog,
struct brw_vertex_program *vp,
const struct brw_vs_prog_key *key,
struct brw_vs_compile *c)
{
memset(c, 0, sizeof(*c));
memcpy(&c->key, key, sizeof(*key));
c->vp = vp;
c->base.shader_prog = prog;
c->base.mem_ctx = ralloc_context(NULL);
}
void
link_varyings::SetUp()
{
this->mem_ctx = ralloc_context(NULL);
this->ir.make_empty();
this->consumer_inputs =
_mesa_hash_table_create(NULL, _mesa_key_hash_string,
_mesa_key_string_equal);
this->consumer_interface_inputs =
_mesa_hash_table_create(NULL, _mesa_key_hash_string,
_mesa_key_string_equal);
}
static void
brw_blorp_params_get_clear_kernel(struct brw_context *brw,
struct brw_blorp_params *params,
bool use_replicated_data)
{
struct brw_blorp_const_color_prog_key blorp_key;
memset(&blorp_key, 0, sizeof(blorp_key));
blorp_key.use_simd16_replicated_data = use_replicated_data;
if (brw_search_cache(&brw->cache, BRW_CACHE_BLORP_PROG,
&blorp_key, sizeof(blorp_key),
¶ms->wm_prog_kernel, ¶ms->wm_prog_data))
return;
void *mem_ctx = ralloc_context(NULL);
nir_builder b;
nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
b.shader->info.name = ralloc_strdup(b.shader, "BLORP-clear");
nir_variable *v_color = nir_variable_create(b.shader, nir_var_shader_in,
glsl_vec4_type(), "v_color");
v_color->data.location = VARYING_SLOT_VAR0;
v_color->data.interpolation = INTERP_MODE_FLAT;
nir_variable *frag_color = nir_variable_create(b.shader, nir_var_shader_out,
glsl_vec4_type(),
"gl_FragColor");
frag_color->data.location = FRAG_RESULT_COLOR;
nir_copy_var(&b, frag_color, v_color);
struct brw_wm_prog_key wm_key;
brw_blorp_init_wm_prog_key(&wm_key);
struct brw_blorp_prog_data prog_data;
unsigned program_size;
const unsigned *program =
brw_blorp_compile_nir_shader(brw, b.shader, &wm_key, use_replicated_data,
&prog_data, &program_size);
brw_upload_cache(&brw->cache, BRW_CACHE_BLORP_PROG,
&blorp_key, sizeof(blorp_key),
program, program_size,
&prog_data, sizeof(prog_data),
¶ms->wm_prog_kernel, ¶ms->wm_prog_data);
ralloc_free(mem_ctx);
}
brw_blorp_const_color_program::brw_blorp_const_color_program(
struct brw_context *brw,
const brw_blorp_const_color_prog_key *key)
: mem_ctx(ralloc_context(NULL)),
brw(brw),
key(key),
R0(),
R1(),
clear_rgba(),
base_mrf(0)
{
prog_data.first_curbe_grf = 0;
prog_data.persample_msaa_dispatch = false;
brw_init_compile(brw, &func, mem_ctx);
}
void
common_builtin::SetUp()
{
this->mem_ctx = ralloc_context(NULL);
this->ir.make_empty();
initialize_context_to_defaults(&this->ctx, API_OPENGL_COMPAT);
this->shader = rzalloc(this->mem_ctx, gl_shader);
this->shader->Type = this->shader_type;
this->shader->Stage = _mesa_shader_enum_to_shader_stage(this->shader_type);
this->state =
new(mem_ctx) _mesa_glsl_parse_state(&this->ctx, this->shader->Stage,
this->shader);
_mesa_glsl_initialize_types(this->state);
_mesa_glsl_initialize_variables(&this->ir, this->state);
}
static void compile_sf_prog( struct brw_context *brw,
struct brw_sf_prog_key *key )
{
const unsigned *program;
void *mem_ctx;
unsigned program_size;
mem_ctx = ralloc_context(NULL);
struct brw_sf_prog_data prog_data;
program = brw_compile_sf(brw->screen->compiler, mem_ctx, key, &prog_data,
&brw->vue_map_geom_out, &program_size);
brw_upload_cache(&brw->cache, BRW_CACHE_SF_PROG,
key, sizeof(*key),
program, program_size,
&prog_data, sizeof(prog_data),
&brw->sf.prog_offset, &brw->sf.prog_data);
ralloc_free(mem_ctx);
}
static bool
split_var_copies_impl(nir_function_impl *impl)
{
struct split_var_copies_state state;
state.mem_ctx = ralloc_parent(impl);
state.dead_ctx = ralloc_context(NULL);
state.progress = false;
nir_foreach_block(impl, split_var_copies_block, &state);
ralloc_free(state.dead_ctx);
if (state.progress) {
nir_metadata_preserve(impl, nir_metadata_block_index |
nir_metadata_dominance);
}
return state.progress;
}
/* Test that we can read and write some large objects, (exercising the code in
* the blob_write functions to realloc blob->data.
*/
static void
test_big_objects(void)
{
void *ctx = ralloc_context(NULL);
struct blob blob;
struct blob_reader reader;
int size = 1000;
int count = 1000;
size_t i;
char *buf;
blob_init(&blob);
/* Initialize our buffer. */
buf = ralloc_size(ctx, size);
for (i = 0; i < size; i++) {
buf[i] = i % 256;
}
/* Write it many times. */
for (i = 0; i < count; i++) {
blob_write_bytes(&blob, buf, size);
}
blob_reader_init(&reader, blob.data, blob.size);
/* Read and verify it many times. */
for (i = 0; i < count; i++) {
expect_equal_bytes((uint8_t *) buf, blob_read_bytes(&reader, size), size,
"read of large objects");
}
expect_equal(reader.end - reader.data, reader.current - reader.data,
"number of bytes read reading large objects");
expect_equal(false, reader.overrun,
"overrun flag not set reading large objects");
blob_finish(&blob);
ralloc_free(ctx);
}
/* Test that we detect overrun. */
static void
test_overrun(void)
{
void *ctx =ralloc_context(NULL);
struct blob *blob;
struct blob_reader reader;
uint32_t value = 0xdeadbeef;
blob = blob_create(ctx);
blob_write_uint32(blob, value);
blob_reader_init(&reader, blob->data, blob->size);
expect_equal(value, blob_read_uint32(&reader), "read before overrun");
expect_equal(false, reader.overrun, "overrun flag not set");
expect_equal(0, blob_read_uint32(&reader), "read at overrun");
expect_equal(true, reader.overrun, "overrun flag set");
ralloc_free(ctx);
}
static bool
brw_codegen_wm_prog(struct brw_context *brw,
struct brw_program *fp,
struct brw_wm_prog_key *key,
struct brw_vue_map *vue_map)
{
const struct gen_device_info *devinfo = &brw->screen->devinfo;
void *mem_ctx = ralloc_context(NULL);
struct brw_wm_prog_data prog_data;
const GLuint *program;
bool start_busy = false;
double start_time = 0;
nir_shader *nir = nir_shader_clone(mem_ctx, fp->program.nir);
memset(&prog_data, 0, sizeof(prog_data));
/* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
if (fp->program.is_arb_asm)
prog_data.base.use_alt_mode = true;
assign_fs_binding_table_offsets(devinfo, &fp->program, key, &prog_data);
if (!fp->program.is_arb_asm) {
brw_nir_setup_glsl_uniforms(mem_ctx, nir, &fp->program,
&prog_data.base, true);
brw_nir_analyze_ubo_ranges(brw->screen->compiler, nir,
NULL, prog_data.base.ubo_ranges);
} else {
brw_nir_setup_arb_uniforms(mem_ctx, nir, &fp->program, &prog_data.base);
if (unlikely(INTEL_DEBUG & DEBUG_WM))
brw_dump_arb_asm("fragment", &fp->program);
}
if (unlikely(brw->perf_debug)) {
start_busy = (brw->batch.last_bo &&
brw_bo_busy(brw->batch.last_bo));
start_time = get_time();
}
int st_index8 = -1, st_index16 = -1, st_index32 = -1;
if (INTEL_DEBUG & DEBUG_SHADER_TIME) {
st_index8 = brw_get_shader_time_index(brw, &fp->program, ST_FS8,
!fp->program.is_arb_asm);
st_index16 = brw_get_shader_time_index(brw, &fp->program, ST_FS16,
!fp->program.is_arb_asm);
st_index32 = brw_get_shader_time_index(brw, &fp->program, ST_FS32,
!fp->program.is_arb_asm);
}
char *error_str = NULL;
program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
key, &prog_data, nir,
&fp->program, st_index8, st_index16, st_index32,
true, false, vue_map,
&error_str);
if (program == NULL) {
if (!fp->program.is_arb_asm) {
fp->program.sh.data->LinkStatus = LINKING_FAILURE;
ralloc_strcat(&fp->program.sh.data->InfoLog, error_str);
}
_mesa_problem(NULL, "Failed to compile fragment shader: %s\n", error_str);
ralloc_free(mem_ctx);
return false;
}
if (unlikely(brw->perf_debug)) {
if (fp->compiled_once) {
brw_debug_recompile(brw, MESA_SHADER_FRAGMENT, fp->program.Id,
key->program_string_id, key);
}
fp->compiled_once = true;
if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
perf_debug("FS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
}
brw_alloc_stage_scratch(brw, &brw->wm.base, prog_data.base.total_scratch);
if (unlikely((INTEL_DEBUG & DEBUG_WM) && fp->program.is_arb_asm))
fprintf(stderr, "\n");
/* The param and pull_param arrays will be freed by the shader cache. */
ralloc_steal(NULL, prog_data.base.param);
ralloc_steal(NULL, prog_data.base.pull_param);
brw_upload_cache(&brw->cache, BRW_CACHE_FS_PROG,
key, sizeof(struct brw_wm_prog_key),
program, prog_data.base.program_size,
&prog_data, sizeof(prog_data),
&brw->wm.base.prog_offset, &brw->wm.base.prog_data);
ralloc_free(mem_ctx);
return true;
}
bool
brw_codegen_gs_prog(struct brw_context *brw,
struct gl_shader_program *prog,
struct brw_geometry_program *gp,
struct brw_gs_prog_key *key)
{
struct brw_stage_state *stage_state = &brw->gs.base;
struct brw_gs_compile c;
memset(&c, 0, sizeof(c));
c.key = *key;
c.gp = gp;
c.prog_data.include_primitive_id =
(gp->program.Base.InputsRead & VARYING_BIT_PRIMITIVE_ID) != 0;
c.prog_data.invocations = gp->program.Invocations;
/* Allocate the references to the uniforms that will end up in the
* prog_data associated with the compiled program, and which will be freed
* by the state cache.
*
* Note: param_count needs to be num_uniform_components * 4, since we add
* padding around uniform values below vec4 size, so the worst case is that
* every uniform is a float which gets padded to the size of a vec4.
*/
struct gl_shader *gs = prog->_LinkedShaders[MESA_SHADER_GEOMETRY];
int param_count = gs->num_uniform_components * 4;
/* We also upload clip plane data as uniforms */
param_count += MAX_CLIP_PLANES * 4;
c.prog_data.base.base.param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
c.prog_data.base.base.pull_param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
c.prog_data.base.base.nr_params = param_count;
if (brw->gen >= 7) {
if (gp->program.OutputType == GL_POINTS) {
/* When the output type is points, the geometry shader may output data
* to multiple streams, and EndPrimitive() has no effect. So we
* configure the hardware to interpret the control data as stream ID.
*/
c.prog_data.control_data_format = GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_SID;
/* We only have to emit control bits if we are using streams */
if (prog->Geom.UsesStreams)
c.control_data_bits_per_vertex = 2;
else
c.control_data_bits_per_vertex = 0;
} else {
/* When the output type is triangle_strip or line_strip, EndPrimitive()
* may be used to terminate the current strip and start a new one
* (similar to primitive restart), and outputting data to multiple
* streams is not supported. So we configure the hardware to interpret
* the control data as EndPrimitive information (a.k.a. "cut bits").
*/
c.prog_data.control_data_format = GEN7_GS_CONTROL_DATA_FORMAT_GSCTL_CUT;
/* We only need to output control data if the shader actually calls
* EndPrimitive().
*/
c.control_data_bits_per_vertex = gp->program.UsesEndPrimitive ? 1 : 0;
}
} else {
/* There are no control data bits in gen6. */
c.control_data_bits_per_vertex = 0;
/* If it is using transform feedback, enable it */
if (prog->TransformFeedback.NumVarying)
c.prog_data.gen6_xfb_enabled = true;
else
c.prog_data.gen6_xfb_enabled = false;
}
c.control_data_header_size_bits =
gp->program.VerticesOut * c.control_data_bits_per_vertex;
/* 1 HWORD = 32 bytes = 256 bits */
c.prog_data.control_data_header_size_hwords =
ALIGN(c.control_data_header_size_bits, 256) / 256;
GLbitfield64 outputs_written = gp->program.Base.OutputsWritten;
/* In order for legacy clipping to work, we need to populate the clip
* distance varying slots whenever clipping is enabled, even if the vertex
* shader doesn't write to gl_ClipDistance.
*/
if (c.key.base.userclip_active) {
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
}
brw_compute_vue_map(brw->intelScreen->devinfo,
&c.prog_data.base.vue_map, outputs_written);
/* Compute the output vertex size.
*
* From the Ivy Bridge PRM, Vol2 Part1 7.2.1.1 STATE_GS - Output Vertex
* Size (p168):
*
* [0,62] indicating [1,63] 16B units
*
* Specifies the size of each vertex stored in the GS output entry
* (following any Control Header data) as a number of 128-bit units
* (minus one).
*
* Programming Restrictions: The vertex size must be programmed as a
* multiple of 32B units with the following exception: Rendering is
* disabled (as per SOL stage state) and the vertex size output by the
* GS thread is 16B.
*
* If rendering is enabled (as per SOL state) the vertex size must be
* programmed as a multiple of 32B units. In other words, the only time
* software can program a vertex size with an odd number of 16B units
* is when rendering is disabled.
*
* Note: B=bytes in the above text.
*
* It doesn't seem worth the extra trouble to optimize the case where the
* vertex size is 16B (especially since this would require special-casing
* the GEN assembly that writes to the URB). So we just set the vertex
* size to a multiple of 32B (2 vec4's) in all cases.
*
* The maximum output vertex size is 62*16 = 992 bytes (31 hwords). We
* budget that as follows:
*
* 512 bytes for varyings (a varying component is 4 bytes and
* gl_MaxGeometryOutputComponents = 128)
* 16 bytes overhead for VARYING_SLOT_PSIZ (each varying slot is 16
* bytes)
* 16 bytes overhead for gl_Position (we allocate it a slot in the VUE
* even if it's not used)
* 32 bytes overhead for gl_ClipDistance (we allocate it 2 VUE slots
* whenever clip planes are enabled, even if the shader doesn't
* write to gl_ClipDistance)
* 16 bytes overhead since the VUE size must be a multiple of 32 bytes
* (see below)--this causes up to 1 VUE slot to be wasted
* 400 bytes available for varying packing overhead
*
* Worst-case varying packing overhead is 3/4 of a varying slot (12 bytes)
* per interpolation type, so this is plenty.
*
*/
unsigned output_vertex_size_bytes = c.prog_data.base.vue_map.num_slots * 16;
assert(brw->gen == 6 ||
output_vertex_size_bytes <= GEN7_MAX_GS_OUTPUT_VERTEX_SIZE_BYTES);
c.prog_data.output_vertex_size_hwords =
ALIGN(output_vertex_size_bytes, 32) / 32;
/* Compute URB entry size. The maximum allowed URB entry size is 32k.
* That divides up as follows:
*
* 64 bytes for the control data header (cut indices or StreamID bits)
* 4096 bytes for varyings (a varying component is 4 bytes and
* gl_MaxGeometryTotalOutputComponents = 1024)
* 4096 bytes overhead for VARYING_SLOT_PSIZ (each varying slot is 16
* bytes/vertex and gl_MaxGeometryOutputVertices is 256)
* 4096 bytes overhead for gl_Position (we allocate it a slot in the VUE
* even if it's not used)
* 8192 bytes overhead for gl_ClipDistance (we allocate it 2 VUE slots
* whenever clip planes are enabled, even if the shader doesn't
* write to gl_ClipDistance)
* 4096 bytes overhead since the VUE size must be a multiple of 32
* bytes (see above)--this causes up to 1 VUE slot to be wasted
* 8128 bytes available for varying packing overhead
*
* Worst-case varying packing overhead is 3/4 of a varying slot per
* interpolation type, which works out to 3072 bytes, so this would allow
* us to accommodate 2 interpolation types without any danger of running
* out of URB space.
*
* In practice, the risk of running out of URB space is very small, since
* the above figures are all worst-case, and most of them scale with the
* number of output vertices. So we'll just calculate the amount of space
* we need, and if it's too large, fail to compile.
*
* The above is for gen7+ where we have a single URB entry that will hold
* all the output. In gen6, we will have to allocate URB entries for every
* vertex we emit, so our URB entries only need to be large enough to hold
* a single vertex. Also, gen6 does not have a control data header.
*/
unsigned output_size_bytes;
if (brw->gen >= 7) {
output_size_bytes =
c.prog_data.output_vertex_size_hwords * 32 * gp->program.VerticesOut;
output_size_bytes += 32 * c.prog_data.control_data_header_size_hwords;
} else {
output_size_bytes = c.prog_data.output_vertex_size_hwords * 32;
}
/* Broadwell stores "Vertex Count" as a full 8 DWord (32 byte) URB output,
* which comes before the control header.
*/
if (brw->gen >= 8)
output_size_bytes += 32;
assert(output_size_bytes >= 1);
int max_output_size_bytes = GEN7_MAX_GS_URB_ENTRY_SIZE_BYTES;
if (brw->gen == 6)
max_output_size_bytes = GEN6_MAX_GS_URB_ENTRY_SIZE_BYTES;
if (output_size_bytes > max_output_size_bytes)
return false;
/* URB entry sizes are stored as a multiple of 64 bytes in gen7+ and
* a multiple of 128 bytes in gen6.
*/
if (brw->gen >= 7)
c.prog_data.base.urb_entry_size = ALIGN(output_size_bytes, 64) / 64;
else
c.prog_data.base.urb_entry_size = ALIGN(output_size_bytes, 128) / 128;
c.prog_data.output_topology =
get_hw_prim_for_gl_prim(gp->program.OutputType);
brw_compute_vue_map(brw->intelScreen->devinfo,
&c.input_vue_map, c.key.input_varyings);
/* GS inputs are read from the VUE 256 bits (2 vec4's) at a time, so we
* need to program a URB read length of ceiling(num_slots / 2).
*/
c.prog_data.base.urb_read_length = (c.input_vue_map.num_slots + 1) / 2;
void *mem_ctx = ralloc_context(NULL);
unsigned program_size;
const unsigned *program =
brw_gs_emit(brw, prog, &c, mem_ctx, &program_size);
if (program == NULL) {
ralloc_free(mem_ctx);
return false;
}
/* Scratch space is used for register spilling */
if (c.base.last_scratch) {
perf_debug("Geometry shader triggered register spilling. "
"Try reducing the number of live vec4 values to "
"improve performance.\n");
c.prog_data.base.base.total_scratch
= brw_get_scratch_size(c.base.last_scratch*REG_SIZE);
brw_get_scratch_bo(brw, &stage_state->scratch_bo,
c.prog_data.base.base.total_scratch *
brw->max_gs_threads);
}
brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
&c.key, sizeof(c.key),
program, program_size,
&c.prog_data, sizeof(c.prog_data),
&stage_state->prog_offset, &brw->gs.prog_data);
ralloc_free(mem_ctx);
return true;
}
bool
brw_codegen_vs_prog(struct brw_context *brw,
struct gl_shader_program *prog,
struct brw_vertex_program *vp,
struct brw_vs_prog_key *key)
{
GLuint program_size;
const GLuint *program;
struct brw_vs_prog_data prog_data;
struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
void *mem_ctx;
int i;
struct brw_shader *vs = NULL;
bool start_busy = false;
double start_time = 0;
if (prog)
vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
memset(&prog_data, 0, sizeof(prog_data));
/* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
if (!prog)
stage_prog_data->use_alt_mode = true;
mem_ctx = ralloc_context(NULL);
brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX,
brw->intelScreen->devinfo,
prog, &vp->program.Base,
&prog_data.base.base, 0);
/* Allocate the references to the uniforms that will end up in the
* prog_data associated with the compiled program, and which will be freed
* by the state cache.
*/
int param_count = vp->program.Base.nir->num_uniforms;
if (!brw->intelScreen->compiler->scalar_vs)
param_count *= 4;
if (vs)
prog_data.base.base.nr_image_params = vs->base.NumImages;
/* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
* planes as uniforms.
*/
param_count += key->nr_userclip_plane_consts * 4;
stage_prog_data->param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
stage_prog_data->pull_param =
rzalloc_array(NULL, const gl_constant_value *, param_count);
stage_prog_data->image_param =
rzalloc_array(NULL, struct brw_image_param,
stage_prog_data->nr_image_params);
stage_prog_data->nr_params = param_count;
if (prog) {
brw_nir_setup_glsl_uniforms(vp->program.Base.nir, prog, &vp->program.Base,
&prog_data.base.base,
brw->intelScreen->compiler->scalar_vs);
} else {
brw_nir_setup_arb_uniforms(vp->program.Base.nir, &vp->program.Base,
&prog_data.base.base);
}
GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
prog_data.inputs_read = vp->program.Base.InputsRead;
if (key->copy_edgeflag) {
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
}
if (brw->gen < 6) {
/* Put dummy slots into the VUE for the SF to put the replaced
* point sprite coords in. We shouldn't need these dummy slots,
* which take up precious URB space, but it would mean that the SF
* doesn't get nice aligned pairs of input coords into output
* coords, which would be a pain to handle.
*/
for (i = 0; i < 8; i++) {
if (key->point_coord_replace & (1 << i))
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
}
/* if back colors are written, allocate slots for front colors too */
if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
}
/* In order for legacy clipping to work, we need to populate the clip
* distance varying slots whenever clipping is enabled, even if the vertex
* shader doesn't write to gl_ClipDistance.
*/
if (key->nr_userclip_plane_consts > 0) {
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
}
brw_compute_vue_map(brw->intelScreen->devinfo,
&prog_data.base.vue_map, outputs_written,
prog ? prog->SeparateShader : false);
if (0) {
_mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
true);
}
if (unlikely(brw->perf_debug)) {
start_busy = (brw->batch.last_bo &&
drm_intel_bo_busy(brw->batch.last_bo));
start_time = get_time();
}
if (unlikely(INTEL_DEBUG & DEBUG_VS))
brw_dump_ir("vertex", prog, vs ? &vs->base : NULL, &vp->program.Base);
int st_index = -1;
if (INTEL_DEBUG & DEBUG_SHADER_TIME)
st_index = brw_get_shader_time_index(brw, prog, &vp->program.Base, ST_VS);
/* Emit GEN4 code.
*/
char *error_str;
program = brw_compile_vs(brw->intelScreen->compiler, brw, mem_ctx, key,
&prog_data, vp->program.Base.nir,
brw_select_clip_planes(&brw->ctx),
!_mesa_is_gles3(&brw->ctx),
st_index, &program_size, &error_str);
if (program == NULL) {
if (prog) {
prog->LinkStatus = false;
ralloc_strcat(&prog->InfoLog, error_str);
}
_mesa_problem(NULL, "Failed to compile vertex shader: %s\n", error_str);
ralloc_free(mem_ctx);
return false;
}
if (unlikely(brw->perf_debug) && vs) {
if (vs->compiled_once) {
brw_vs_debug_recompile(brw, prog, key);
}
if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
perf_debug("VS compile took %.03f ms and stalled the GPU\n",
(get_time() - start_time) * 1000);
}
vs->compiled_once = true;
}
/* Scratch space is used for register spilling */
if (prog_data.base.base.total_scratch) {
brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
prog_data.base.base.total_scratch *
brw->max_vs_threads);
}
brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
key, sizeof(struct brw_vs_prog_key),
program, program_size,
&prog_data, sizeof(prog_data),
&brw->vs.base.prog_offset, &brw->vs.prog_data);
ralloc_free(mem_ctx);
return true;
}
glslopt_ctx (glslopt_target target) {
mem_ctx = ralloc_context (NULL);
initialize_mesa_context (&mesa_ctx, target);
}
bool
fs_visitor::opt_cse_local(bblock_t *block, exec_list *aeb)
{
bool progress = false;
void *mem_ctx = ralloc_context(this->mem_ctx);
for (fs_inst *inst = (fs_inst *)block->start;
inst != block->end->next;
inst = (fs_inst *) inst->next) {
/* Skip some cases. */
if (is_expression(inst) && !inst->predicate && inst->mlen == 0 &&
!inst->force_uncompressed && !inst->force_sechalf &&
!inst->conditional_mod)
{
bool found = false;
aeb_entry *entry;
foreach_list(entry_node, aeb) {
entry = (aeb_entry *) entry_node;
/* Match current instruction's expression against those in AEB. */
if (inst->opcode == entry->generator->opcode &&
inst->saturate == entry->generator->saturate &&
operands_match(entry->generator->src, inst->src)) {
found = true;
progress = true;
break;
}
}
if (!found) {
/* Our first sighting of this expression. Create an entry. */
aeb_entry *entry = ralloc(mem_ctx, aeb_entry);
entry->tmp = reg_undef;
entry->generator = inst;
aeb->push_tail(entry);
} else {
/* This is at least our second sighting of this expression.
* If we don't have a temporary already, make one.
*/
bool no_existing_temp = entry->tmp.file == BAD_FILE;
if (no_existing_temp) {
entry->tmp = fs_reg(this, glsl_type::float_type);
entry->tmp.type = inst->dst.type;
fs_inst *copy = new(ralloc_parent(inst))
fs_inst(BRW_OPCODE_MOV, entry->generator->dst, entry->tmp);
entry->generator->insert_after(copy);
entry->generator->dst = entry->tmp;
}
/* dest <- temp */
fs_inst *copy = new(ralloc_parent(inst))
fs_inst(BRW_OPCODE_MOV, inst->dst, entry->tmp);
inst->replace_with(copy);
/* Appending an instruction may have changed our bblock end. */
if (inst == block->end) {
block->end = copy;
}
/* Continue iteration with copy->next */
inst = copy;
}
}
int main(int argc, char **argv)
{
int i;
Light light;
const char *filename;
Camera cam;
Vec3 position = {0, 0, 150e9};
Vec3 up = {0, 1, 0};
Vec3 target = {0, 0, 0};
Shader *shader_light;
Shader *shader_simple;
Mesh *mesh;
Renderable planet;
if (argc < 2)
filename = STRINGIFY(ROOT_PATH) "/data/teapot.ply";
else
filename = argv[1];
solsys = solsys_load(STRINGIFY(ROOT_PATH) "/data/sol.ini");
if (solsys == NULL)
return 1;
mesh = mesh_import(filename);
if (mesh == NULL)
return 1;
for (i = 0; i < mesh->num_vertices; i++) /* Blow up the teapot */
{
mesh->vertex[i].x = mesh->vertex[i].x * 100;
mesh->vertex[i].y = mesh->vertex[i].y * 100;
mesh->vertex[i].z = mesh->vertex[i].z * 100;
}
cam.fov = M_PI/4;
cam.left = 0;
cam.bottom = 0;
cam.width = 1024;
cam.height = 768;
cam.zNear = 1e6;
cam.zFar = 4.5e15;
init_allegro(&cam);
cam_lookat(&cam, position, target, up);
glewInit();
shader_light = shader_create(STRINGIFY(ROOT_PATH) "/data/lighting.v.glsl",
STRINGIFY(ROOT_PATH) "/data/lighting.f.glsl");
if (shader_light == NULL)
return 1;
shader_simple = shader_create(STRINGIFY(ROOT_PATH) "/data/simple.v.glsl",
STRINGIFY(ROOT_PATH) "/data/simple.f.glsl");
if (shader_simple == NULL)
return 1;
glmProjectionMatrix = glmNewMatrixStack();
glmViewMatrix = glmNewMatrixStack();
glmModelMatrix = glmNewMatrixStack();
light.position = light_pos;
memcpy(light.ambient, light_ambient, sizeof(light_ambient));
memcpy(light.diffuse, light_diffuse, sizeof(light_diffuse));
memcpy(light.specular, light_specular, sizeof(light_specular));
glClearColor(20/255., 30/255., 50/255., 1.0);
glEnable(GL_DEPTH_TEST);
glEnable(GL_CULL_FACE);
glCullFace(GL_BACK);
glPointSize(2);
planet.data = mesh;
planet.upload_to_gpu = mesh_upload_to_gpu;
planet.render = mesh_render;
planet.shader = shader_light;
renderable_upload_to_gpu(&planet);
/* Transformation matrices */
cam_projection_matrix(&cam, glmProjectionMatrix);
/* Start rendering */
while(handle_input(ev_queue, &cam))
{
void *ctx;
Entity *renderlist, *prev;
t += 365*86400;
/* Physics stuff */
solsys_update(solsys, t);
ctx = ralloc_context(NULL);
prev = NULL;
for (i = 0; i < solsys->num_bodies; i++)
{
Entity *e;
e = ralloc(ctx, Entity);
e->orientation = (Quaternion) {1, 0, 0, 0};
e->renderable = &planet;
e->position = solsys->body[i].position;
e->radius = solsys->body[i].radius;
e->prev = prev;
e->next = NULL;
if (prev != NULL)
e->prev->next = e;
prev = e;
if (i == 0)
renderlist = e;
}
/* Rendering stuff */
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
glmLoadIdentity(glmModelMatrix);
glmLoadIdentity(glmViewMatrix);
cam_view_matrix(&cam, glmViewMatrix); /* view */
light_upload_to_gpu(&light, shader_light);
render_entity_list(renderlist);
al_flip_display();
calcfps();
ralloc_free(ctx);
}
ralloc_free(mesh);
ralloc_free(solsys);
shader_delete(shader_light);
shader_delete(shader_simple);
glmFreeMatrixStack(glmProjectionMatrix);
glmFreeMatrixStack(glmViewMatrix);
glmFreeMatrixStack(glmModelMatrix);
al_destroy_display(dpy);
return 0;
}
