/**
* Create new draw module context with gallivm state for LLVM JIT.
*/
struct draw_context *
draw_create_gallivm(struct pipe_context *pipe, struct gallivm_state *gallivm)
{
struct draw_context *draw = CALLOC_STRUCT( draw_context );
if (draw == NULL)
goto fail;
#if HAVE_LLVM
if (draw_get_option_use_llvm()) {
if (!gallivm) {
gallivm = gallivm_create();
draw->own_gallivm = gallivm;
}
if (gallivm)
draw->llvm = draw_llvm_create(draw, gallivm);
}
#endif
if (!draw_init(draw))
goto fail;
draw->pipe = pipe;
return draw;
fail:
draw_destroy( draw );
return NULL;
}
/**
* Create new draw module context with gallivm state for LLVM JIT.
*/
static struct draw_context *
draw_create_context(struct pipe_context *pipe, boolean try_llvm)
{
struct draw_context *draw = CALLOC_STRUCT( draw_context );
if (draw == NULL)
goto err_out;
#if HAVE_LLVM
if (try_llvm && draw_get_option_use_llvm()) {
draw->llvm = draw_llvm_create(draw);
if (!draw->llvm)
goto err_destroy;
}
#endif
draw->pipe = pipe;
if (!draw_init(draw))
goto err_destroy;
return draw;
err_destroy:
draw_destroy( draw );
err_out:
return NULL;
}
/**
* Create new draw module context with gallivm state for LLVM JIT.
*/
static struct draw_context *
draw_create_context(struct pipe_context *pipe, void *context,
boolean try_llvm)
{
struct draw_context *draw = CALLOC_STRUCT( draw_context );
if (!draw)
goto err_out;
/* we need correct cpu caps for disabling denorms in draw_vbo() */
util_cpu_detect();
#if HAVE_LLVM
if (try_llvm && draw_get_option_use_llvm()) {
draw->llvm = draw_llvm_create(draw, (LLVMContextRef)context);
}
#endif
draw->pipe = pipe;
if (!draw_init(draw))
goto err_destroy;
draw->ia = draw_prim_assembler_create(draw);
if (!draw->ia)
goto err_destroy;
return draw;
err_destroy:
draw_destroy( draw );
err_out:
return NULL;
}
void draw_geometry_shader_prepare(struct draw_geometry_shader *shader,
struct draw_context *draw)
{
#ifdef HAVE_LLVM
boolean use_llvm = draw_get_option_use_llvm();
#else
boolean use_llvm = FALSE;
#endif
if (!use_llvm && shader && shader->machine->Tokens != shader->state.tokens) {
tgsi_exec_machine_bind_shader(shader->machine,
shader->state.tokens,
draw->gs.tgsi.sampler);
}
}
/**
* XXX: Results for PIPE_SHADER_CAP_MAX_TEXTURE_SAMPLERS because there are two
* different ways of setting textures, and drivers typically only support one.
*/
int
draw_get_shader_param(unsigned shader, enum pipe_shader_cap param)
{
#ifdef HAVE_LLVM
if (draw_get_option_use_llvm()) {
switch(shader) {
case PIPE_SHADER_VERTEX:
case PIPE_SHADER_GEOMETRY:
return gallivm_get_shader_param(param);
default:
return 0;
}
}
#endif
return draw_get_shader_param_no_llvm(shader, param);
}
void draw_delete_geometry_shader(struct draw_context *draw,
struct draw_geometry_shader *dgs)
{
if (!dgs) {
return;
}
#ifdef HAVE_LLVM
if (draw_get_option_use_llvm()) {
struct llvm_geometry_shader *shader = llvm_geometry_shader(dgs);
struct draw_gs_llvm_variant_list_item *li;
li = first_elem(&shader->variants);
while(!at_end(&shader->variants, li)) {
struct draw_gs_llvm_variant_list_item *next = next_elem(li);
draw_gs_llvm_destroy_variant(li->base);
li = next;
}
assert(shader->variants_cached == 0);
if (dgs->llvm_prim_lengths) {
unsigned i;
for (i = 0; i < dgs->max_out_prims; ++i) {
align_free(dgs->llvm_prim_lengths[i]);
}
FREE(dgs->llvm_prim_lengths);
}
align_free(dgs->llvm_emitted_primitives);
align_free(dgs->llvm_emitted_vertices);
align_free(dgs->llvm_prim_ids);
align_free(dgs->gs_input);
}
#endif
FREE(dgs->primitive_lengths);
FREE((void*) dgs->state.tokens);
FREE(dgs);
}
struct draw_geometry_shader *
draw_create_geometry_shader(struct draw_context *draw,
const struct pipe_shader_state *state)
{
#ifdef HAVE_LLVM
boolean use_llvm = draw_get_option_use_llvm();
struct llvm_geometry_shader *llvm_gs;
#endif
struct draw_geometry_shader *gs;
unsigned i;
#ifdef HAVE_LLVM
if (use_llvm) {
llvm_gs = CALLOC_STRUCT(llvm_geometry_shader);
if (llvm_gs == NULL)
return NULL;
gs = &llvm_gs->base;
make_empty_list(&llvm_gs->variants);
} else
#endif
{
gs = CALLOC_STRUCT(draw_geometry_shader);
}
if (!gs)
return NULL;
gs->draw = draw;
gs->state = *state;
gs->state.tokens = tgsi_dup_tokens(state->tokens);
if (!gs->state.tokens) {
FREE(gs);
return NULL;
}
tgsi_scan_shader(state->tokens, &gs->info);
/* setup the defaults */
gs->input_primitive = PIPE_PRIM_TRIANGLES;
gs->output_primitive = PIPE_PRIM_TRIANGLE_STRIP;
gs->max_output_vertices = 32;
gs->max_out_prims = 0;
#ifdef HAVE_LLVM
if (use_llvm) {
/* TODO: change the input array to handle the following
vector length, instead of the currently hardcoded
TGSI_NUM_CHANNELS
gs->vector_length = lp_native_vector_width / 32;*/
gs->vector_length = TGSI_NUM_CHANNELS;
} else
#endif
{
gs->vector_length = 1;
}
for (i = 0; i < gs->info.num_properties; ++i) {
if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_INPUT_PRIM)
gs->input_primitive = gs->info.properties[i].data[0];
else if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_OUTPUT_PRIM)
gs->output_primitive = gs->info.properties[i].data[0];
else if (gs->info.properties[i].name ==
TGSI_PROPERTY_GS_MAX_OUTPUT_VERTICES)
gs->max_output_vertices = gs->info.properties[i].data[0];
}
/* Primitive boundary is bigger than max_output_vertices by one, because
* the specification says that the geometry shader should exit if the
* number of emitted vertices is bigger or equal to max_output_vertices and
* we can't do that because we're running in the SoA mode, which means that
* our storing routines will keep getting called on channels that have
* overflown.
* So we need some scratch area where we can keep writing the overflown
* vertices without overwriting anything important or crashing.
*/
gs->primitive_boundary = gs->max_output_vertices + 1;
for (i = 0; i < gs->info.num_outputs; i++) {
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
gs->info.output_semantic_index[i] == 0)
gs->position_output = i;
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_VIEWPORT_INDEX)
gs->viewport_index_output = i;
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_CLIPDIST) {
debug_assert(gs->info.output_semantic_index[i] <
PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT);
gs->clipdistance_output[gs->info.output_semantic_index[i]] = i;
}
if (gs->info.output_semantic_name[i] == TGSI_SEMANTIC_CULLDIST) {
debug_assert(gs->info.output_semantic_index[i] <
PIPE_MAX_CLIP_OR_CULL_DISTANCE_ELEMENT_COUNT);
gs->culldistance_output[gs->info.output_semantic_index[i]] = i;
}
}
gs->machine = draw->gs.tgsi.machine;
#ifdef HAVE_LLVM
if (use_llvm) {
int vector_size = gs->vector_length * sizeof(float);
gs->gs_input = align_malloc(sizeof(struct draw_gs_inputs), 16);
memset(gs->gs_input, 0, sizeof(struct draw_gs_inputs));
gs->llvm_prim_lengths = 0;
gs->llvm_emitted_primitives = align_malloc(vector_size, vector_size);
gs->llvm_emitted_vertices = align_malloc(vector_size, vector_size);
gs->llvm_prim_ids = align_malloc(vector_size, vector_size);
gs->fetch_outputs = llvm_fetch_gs_outputs;
gs->fetch_inputs = llvm_fetch_gs_input;
gs->prepare = llvm_gs_prepare;
gs->run = llvm_gs_run;
gs->jit_context = &draw->llvm->gs_jit_context;
llvm_gs->variant_key_size =
draw_gs_llvm_variant_key_size(
MAX2(gs->info.file_max[TGSI_FILE_SAMPLER]+1,
gs->info.file_max[TGSI_FILE_SAMPLER_VIEW]+1));
} else
#endif
{
gs->fetch_outputs = tgsi_fetch_gs_outputs;
gs->fetch_inputs = tgsi_fetch_gs_input;
gs->prepare = tgsi_gs_prepare;
gs->run = tgsi_gs_run;
}
return gs;
}
/**
* Execute geometry shader.
*/
int draw_geometry_shader_run(struct draw_geometry_shader *shader,
const void *constants[PIPE_MAX_CONSTANT_BUFFERS],
const unsigned constants_size[PIPE_MAX_CONSTANT_BUFFERS],
const struct draw_vertex_info *input_verts,
const struct draw_prim_info *input_prim,
const struct tgsi_shader_info *input_info,
struct draw_vertex_info *output_verts,
struct draw_prim_info *output_prims )
{
const float (*input)[4] = (const float (*)[4])input_verts->verts->data;
unsigned input_stride = input_verts->vertex_size;
unsigned num_outputs = draw_total_gs_outputs(shader->draw);
unsigned vertex_size = sizeof(struct vertex_header) + num_outputs * 4 * sizeof(float);
unsigned num_input_verts = input_prim->linear ?
input_verts->count :
input_prim->count;
unsigned num_in_primitives =
align(
MAX2(u_decomposed_prims_for_vertices(input_prim->prim,
num_input_verts),
u_decomposed_prims_for_vertices(shader->input_primitive,
num_input_verts)),
shader->vector_length);
unsigned max_out_prims =
u_decomposed_prims_for_vertices(shader->output_primitive,
shader->max_output_vertices)
* num_in_primitives;
//Assume at least one primitive
max_out_prims = MAX2(max_out_prims, 1);
output_verts->vertex_size = vertex_size;
output_verts->stride = output_verts->vertex_size;
/* we allocate exactly one extra vertex per primitive to allow the GS to emit
* overflown vertices into some area where they won't harm anyone */
output_verts->verts =
(struct vertex_header *)MALLOC(output_verts->vertex_size *
max_out_prims *
shader->primitive_boundary);
#if 0
debug_printf("%s count = %d (in prims # = %d)\n",
__FUNCTION__, num_input_verts, num_in_primitives);
debug_printf("\tlinear = %d, prim_info->count = %d\n",
input_prim->linear, input_prim->count);
debug_printf("\tprim pipe = %s, shader in = %s, shader out = %s, max out = %d\n",
u_prim_name(input_prim->prim),
u_prim_name(shader->input_primitive),
u_prim_name(shader->output_primitive),
shader->max_output_vertices);
#endif
shader->emitted_vertices = 0;
shader->emitted_primitives = 0;
shader->vertex_size = vertex_size;
shader->tmp_output = (float (*)[4])output_verts->verts->data;
shader->fetched_prim_count = 0;
shader->input_vertex_stride = input_stride;
shader->input = input;
shader->input_info = input_info;
FREE(shader->primitive_lengths);
shader->primitive_lengths = MALLOC(max_out_prims * sizeof(unsigned));
#ifdef HAVE_LLVM
if (draw_get_option_use_llvm()) {
shader->gs_output = output_verts->verts;
if (max_out_prims > shader->max_out_prims) {
unsigned i;
if (shader->llvm_prim_lengths) {
for (i = 0; i < shader->max_out_prims; ++i) {
align_free(shader->llvm_prim_lengths[i]);
}
FREE(shader->llvm_prim_lengths);
}
shader->llvm_prim_lengths = MALLOC(max_out_prims * sizeof(unsigned*));
for (i = 0; i < max_out_prims; ++i) {
int vector_size = shader->vector_length * sizeof(unsigned);
shader->llvm_prim_lengths[i] =
align_malloc(vector_size, vector_size);
}
shader->max_out_prims = max_out_prims;
}
shader->jit_context->prim_lengths = shader->llvm_prim_lengths;
shader->jit_context->emitted_vertices = shader->llvm_emitted_vertices;
shader->jit_context->emitted_prims = shader->llvm_emitted_primitives;
}
#endif
shader->prepare(shader, constants, constants_size);
if (input_prim->linear)
gs_run(shader, input_prim, input_verts,
output_prims, output_verts);
else
gs_run_elts(shader, input_prim, input_verts,
output_prims, output_verts);
/* Flush the remaining primitives. Will happen if
* num_input_primitives % 4 != 0
*/
if (shader->fetched_prim_count > 0) {
gs_flush(shader);
}
debug_assert(shader->fetched_prim_count == 0);
/* Update prim_info:
*/
output_prims->linear = TRUE;
output_prims->elts = NULL;
output_prims->start = 0;
output_prims->count = shader->emitted_vertices;
output_prims->prim = shader->output_primitive;
output_prims->flags = 0x0;
output_prims->primitive_lengths = shader->primitive_lengths;
output_prims->primitive_count = shader->emitted_primitives;
output_verts->count = shader->emitted_vertices;
if (shader->draw->collect_statistics) {
unsigned i;
for (i = 0; i < shader->emitted_primitives; ++i) {
shader->draw->statistics.gs_primitives +=
u_decomposed_prims_for_vertices(shader->output_primitive,
shader->primitive_lengths[i]);
}
}
#if 0
debug_printf("GS finished, prims = %d, verts = %d\n",
output_prims->primitive_count,
output_verts->count);
#endif
return shader->emitted_vertices;
}
