#include "draw_private.h"
#include "draw_context.h"
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
#include "draw_llvm.h"
#endif
#include "tgsi/tgsi_parse.h"
#include "tgsi/tgsi_exec.h"
-
+#include "nir/nir_to_tgsi_info.h"
#include "pipe/p_shader_tokens.h"
#include "util/u_math.h"
#include "util/u_memory.h"
#include "util/u_prim.h"
-
+#include "util/ralloc.h"
/* fixme: move it from here */
#define MAX_PRIMITIVES 64
static inline boolean
draw_gs_should_flush(struct draw_geometry_shader *shader)
{
- return (shader->fetched_prim_count == shader->vector_length);
+ return (shader->fetched_prim_count == shader->vector_length || shader->num_invocations > 1);
}
/*#define DEBUG_OUTPUTS 1*/
static void
tgsi_fetch_gs_outputs(struct draw_geometry_shader *shader,
+ unsigned stream,
unsigned num_primitives,
float (**p_output)[4])
{
struct tgsi_exec_machine *machine = shader->machine;
unsigned prim_idx, j, slot;
- unsigned current_idx = 0;
float (*output)[4];
output = *p_output;
*/
for (prim_idx = 0; prim_idx < num_primitives; ++prim_idx) {
- unsigned num_verts_per_prim = machine->Primitives[prim_idx];
- shader->primitive_lengths[prim_idx + shader->emitted_primitives] =
- machine->Primitives[prim_idx];
- shader->emitted_vertices += num_verts_per_prim;
- for (j = 0; j < num_verts_per_prim; j++, current_idx++) {
- int idx = current_idx * shader->info.num_outputs;
+ unsigned num_verts_per_prim = machine->Primitives[stream][prim_idx];
+ unsigned prim_offset = machine->PrimitiveOffsets[stream][prim_idx];
+ shader->stream[stream].primitive_lengths[prim_idx + shader->stream[stream].emitted_primitives] =
+ machine->Primitives[stream][prim_idx];
+ shader->stream[stream].emitted_vertices += num_verts_per_prim;
+
+ for (j = 0; j < num_verts_per_prim; j++) {
+ int idx = prim_offset + j * shader->info.num_outputs;
#ifdef DEBUG_OUTPUTS
- debug_printf("%d) Output vert:\n", idx / shader->info.num_outputs);
+ debug_printf("%d/%d) Output vert:\n", stream, idx / shader->info.num_outputs);
#endif
for (slot = 0; slot < shader->info.num_outputs; slot++) {
output[slot][0] = machine->Outputs[idx + slot].xyzw[0].f[0];
}
}
*p_output = output;
- shader->emitted_primitives += num_primitives;
+ shader->stream[stream].emitted_primitives += num_primitives;
}
/*#define DEBUG_INPUTS 1*/
unsigned input_vertex_stride = shader->input_vertex_stride;
const float (*input_ptr)[4];
+ int primid_sv = machine->SysSemanticToIndex[TGSI_SEMANTIC_PRIMID];
+ if (primid_sv != -1) {
+ for (unsigned j = 0; j < TGSI_QUAD_SIZE; j++)
+ machine->SystemValue[primid_sv].xyzw[0].i[j] = shader->in_prim_idx;
+ }
+
input_ptr = shader->input;
for (i = 0; i < num_vertices; ++i) {
const unsigned constants_size[PIPE_MAX_CONSTANT_BUFFERS])
{
struct tgsi_exec_machine *machine = shader->machine;
- int j;
tgsi_exec_set_constant_buffers(machine, PIPE_MAX_CONSTANT_BUFFERS,
constants, constants_size);
+}
+
+static void tgsi_gs_run(struct draw_geometry_shader *shader,
+ unsigned input_primitives,
+ unsigned *out_prims)
+{
+ struct tgsi_exec_machine *machine = shader->machine;
+ int i;
if (shader->info.uses_invocationid) {
unsigned i = machine->SysSemanticToIndex[TGSI_SEMANTIC_INVOCATIONID];
- for (j = 0; j < TGSI_QUAD_SIZE; j++)
+ for (int j = 0; j < TGSI_QUAD_SIZE; j++)
machine->SystemValue[i].xyzw[0].i[j] = shader->invocation_id;
}
-}
-
-static unsigned tgsi_gs_run(struct draw_geometry_shader *shader,
- unsigned input_primitives)
-{
- struct tgsi_exec_machine *machine = shader->machine;
/* run interpreter */
tgsi_exec_machine_run(machine, 0);
- return
- machine->Temps[TGSI_EXEC_TEMP_PRIMITIVE_I].xyzw[TGSI_EXEC_TEMP_PRIMITIVE_C].u[0];
+ for (i = 0; i < 4; i++) {
+ int prim_i;
+ int prim_c;
+ switch (i) {
+ case 0:
+ prim_i = TGSI_EXEC_TEMP_PRIMITIVE_I;
+ prim_c = TGSI_EXEC_TEMP_PRIMITIVE_C;
+ break;
+ case 1:
+ prim_i = TGSI_EXEC_TEMP_PRIMITIVE_S1_I;
+ prim_c = TGSI_EXEC_TEMP_PRIMITIVE_S1_C;
+ break;
+ case 2:
+ prim_i = TGSI_EXEC_TEMP_PRIMITIVE_S2_I;
+ prim_c = TGSI_EXEC_TEMP_PRIMITIVE_S2_C;
+ break;
+ case 3:
+ prim_i = TGSI_EXEC_TEMP_PRIMITIVE_S3_I;
+ prim_c = TGSI_EXEC_TEMP_PRIMITIVE_S3_C;
+ break;
+ };
+
+ out_prims[i] = machine->Temps[prim_i].xyzw[prim_c].u[0];
+ }
}
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
static void
llvm_fetch_gs_input(struct draw_geometry_shader *shader,
static void
llvm_fetch_gs_outputs(struct draw_geometry_shader *shader,
+ unsigned stream,
unsigned num_primitives,
float (**p_output)[4])
{
int vertex_count = 0;
int total_prims = 0;
int max_prims_per_invocation = 0;
- char *output_ptr = (char*)shader->gs_output;
+ char *output_ptr = (char*)shader->gs_output[stream];
int i, j, prim_idx;
unsigned next_prim_boundary = shader->primitive_boundary;
for (i = 0; i < shader->vector_length; ++i) {
- int prims = shader->llvm_emitted_primitives[i];
+ int prims = shader->llvm_emitted_primitives[i + (stream * shader->vector_length)];
total_prims += prims;
max_prims_per_invocation = MAX2(max_prims_per_invocation, prims);
}
for (i = 0; i < shader->vector_length; ++i) {
- total_verts += shader->llvm_emitted_vertices[i];
+ total_verts += shader->llvm_emitted_vertices[i + (stream * shader->vector_length)];
}
- output_ptr += shader->emitted_vertices * shader->vertex_size;
+ output_ptr += shader->stream[stream].emitted_vertices * shader->vertex_size;
for (i = 0; i < shader->vector_length - 1; ++i) {
- int current_verts = shader->llvm_emitted_vertices[i];
- int next_verts = shader->llvm_emitted_vertices[i + 1];
+ int current_verts = shader->llvm_emitted_vertices[i + (stream * shader->vector_length)];
+ int next_verts = shader->llvm_emitted_vertices[i + 1 + (stream * shader->vector_length)];
#if 0
int j;
for (j = 0; j < current_verts; ++j) {
prim_idx = 0;
for (i = 0; i < shader->vector_length; ++i) {
- int num_prims = shader->llvm_emitted_primitives[i];
+ int num_prims = shader->llvm_emitted_primitives[i + (stream * shader->vector_length)];
for (j = 0; j < num_prims; ++j) {
int prim_length =
- shader->llvm_prim_lengths[j][i];
- shader->primitive_lengths[shader->emitted_primitives + prim_idx] =
+ shader->llvm_prim_lengths[j * shader->num_vertex_streams + stream][i];
+ shader->stream[stream].primitive_lengths[shader->stream[stream].emitted_primitives + prim_idx] =
prim_length;
++prim_idx;
}
}
- shader->emitted_primitives += total_prims;
- shader->emitted_vertices += total_verts;
+ shader->stream[stream].emitted_primitives += total_prims;
+ shader->stream[stream].emitted_vertices += total_verts;
}
static void
{
}
-static unsigned
+static void
llvm_gs_run(struct draw_geometry_shader *shader,
- unsigned input_primitives)
+ unsigned input_primitives, unsigned *out_prims)
{
- unsigned ret;
- char *input = (char*)shader->gs_output;
-
- input += (shader->emitted_vertices * shader->vertex_size);
+ struct vertex_header *input[PIPE_MAX_VERTEX_STREAMS];
+ for (unsigned i = 0; i < shader->num_vertex_streams; i++) {
+ char *tmp = (char *)shader->gs_output[i];
+ tmp += shader->stream[i].emitted_vertices * shader->vertex_size;
+ input[i] = (struct vertex_header *)tmp;
+ }
- ret = shader->current_variant->jit_func(
+ shader->current_variant->jit_func(
shader->jit_context, shader->gs_input->data,
- (struct vertex_header*)input,
+ input,
input_primitives,
shader->draw->instance_id,
shader->llvm_prim_ids,
shader->invocation_id);
- return ret;
+ for (unsigned i = 0; i < shader->num_vertex_streams; i++) {
+ out_prims[i] = shader->jit_context->emitted_prims[i];
+ }
}
#endif
static void gs_flush(struct draw_geometry_shader *shader)
{
- unsigned out_prim_count;
-
+ unsigned out_prim_count[TGSI_MAX_VERTEX_STREAMS];
+ unsigned i;
unsigned input_primitives = shader->fetched_prim_count;
if (shader->draw->collect_statistics) {
debug_assert(input_primitives > 0 &&
input_primitives <= 4);
- out_prim_count = shader->run(shader, input_primitives);
- shader->fetch_outputs(shader, out_prim_count,
- &shader->tmp_output);
+ for (unsigned invocation = 0; invocation < shader->num_invocations; invocation++) {
+ shader->invocation_id = invocation;
+ shader->run(shader, input_primitives, out_prim_count);
+ for (i = 0; i < shader->num_vertex_streams; i++) {
+ shader->fetch_outputs(shader, i, out_prim_count[i],
+ &shader->stream[i].tmp_output);
+ }
+ }
#if 0
- debug_printf("PRIM emitted prims = %d (verts=%d), cur prim count = %d\n",
- shader->emitted_primitives, shader->emitted_vertices,
- out_prim_count);
+ for (i = 0; i < shader->num_vertex_streams; i++) {
+ debug_printf("stream %d: PRIM emitted prims = %d (verts=%d), cur prim count = %d\n",
+ i,
+ shader->stream[i].emitted_primitives, shader->stream[i].emitted_vertices,
+ out_prim_count[i]);
+ }
#endif
shader->fetched_prim_count = 0;
* overflown vertices into some area where they won't harm anyone */
unsigned total_verts_per_buffer = shader->primitive_boundary *
num_in_primitives;
- unsigned invocation;
+ int i;
//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;
- output_verts->verts =
- (struct vertex_header *)MALLOC(output_verts->vertex_size *
- total_verts_per_buffer * shader->num_invocations);
- debug_assert(output_verts->verts);
+ for (i = 0; i < shader->num_vertex_streams; i++) {
+ /* write all the vertex data into all the streams */
+ output_verts[i].vertex_size = vertex_size;
+ output_verts[i].stride = output_verts[i].vertex_size;
+ output_verts[i].verts =
+ (struct vertex_header *)MALLOC(output_verts[i].vertex_size *
+ total_verts_per_buffer * shader->num_invocations);
+ debug_assert(output_verts[i].verts);
+ }
#if 0
- debug_printf("%s count = %d (in prims # = %d)\n",
- __FUNCTION__, num_input_verts, num_in_primitives);
+ debug_printf("%s count = %d (in prims # = %d, invocs = %d, streams = %d)\n",
+ __FUNCTION__, num_input_verts, num_in_primitives,
+ shader->num_invocations, shader->num_vertex_streams);
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\n"
+ debug_printf("\tprim pipe = %s, shader in = %s, shader out = %s\n",
u_prim_name(input_prim->prim),
u_prim_name(shader->input_primitive),
u_prim_name(shader->output_primitive));
total_verts_per_buffer);
#endif
- shader->emitted_vertices = 0;
- shader->emitted_primitives = 0;
+ for (i = 0; i < shader->num_vertex_streams; i++) {
+ shader->stream[i].emitted_vertices = 0;
+ shader->stream[i].emitted_primitives = 0;
+ FREE(shader->stream[i].primitive_lengths);
+ shader->stream[i].primitive_lengths = MALLOC(max_out_prims * sizeof(unsigned) * shader->num_invocations);
+ shader->stream[i].tmp_output = (float (*)[4])output_verts[i].verts->data;
+ }
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) * shader->num_invocations);
-
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
if (shader->draw->llvm) {
- shader->gs_output = output_verts->verts;
+ for (i = 0; i < shader->num_vertex_streams; i++) {
+ shader->gs_output[i] = output_verts[i].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) {
+ for (i = 0; i < shader->num_vertex_streams * 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) {
+ shader->llvm_prim_lengths = MALLOC(shader->num_vertex_streams * max_out_prims * sizeof(unsigned*));
+ for (i = 0; i < shader->num_vertex_streams * max_out_prims; ++i) {
int vector_size = shader->vector_length * sizeof(unsigned);
shader->llvm_prim_lengths[i] =
align_malloc(vector_size, vector_size);
}
#endif
- for (invocation = 0; invocation < shader->num_invocations; invocation++) {
- shader->invocation_id = invocation;
-
- shader->prepare(shader, constants, constants_size);
+ 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);
+ 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);
+ /* 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]);
+ for (i = 0; i < shader->num_vertex_streams; i++) {
+ output_prims[i].linear = TRUE;
+ output_prims[i].elts = NULL;
+ output_prims[i].start = 0;
+ output_prims[i].count = shader->stream[i].emitted_vertices;
+ output_prims[i].prim = shader->output_primitive;
+ output_prims[i].flags = 0x0;
+ output_prims[i].primitive_lengths = shader->stream[i].primitive_lengths;
+ output_prims[i].primitive_count = shader->stream[i].emitted_primitives;
+ output_verts[i].count = shader->stream[i].emitted_vertices;
+
+ if (shader->draw->collect_statistics) {
+ unsigned j;
+ for (j = 0; j < shader->stream[i].emitted_primitives; ++j) {
+ shader->draw->statistics.gs_primitives +=
+ u_decomposed_prims_for_vertices(shader->output_primitive,
+ shader->stream[i].primitive_lengths[j]);
+ }
}
}
#if 0
- debug_printf("GS finished, prims = %d, verts = %d\n",
- output_prims->primitive_count,
- output_verts->count);
+ debug_printf("GS finished\n");
+ for (i = 0; i < 4; i++)
+ debug_printf("stream %d: prims = %d verts = %d\n", i, output_prims[i].primitive_count, output_verts[i].count);
#endif
- return shader->emitted_vertices;
+ return 0;
}
void draw_geometry_shader_prepare(struct draw_geometry_shader *shader,
{
if (!draw->llvm) {
draw->gs.tgsi.machine = tgsi_exec_machine_create(PIPE_SHADER_GEOMETRY);
- if (!draw->gs.tgsi.machine)
- return FALSE;
- draw->gs.tgsi.machine->Primitives = align_malloc(
- MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector), 16);
- if (!draw->gs.tgsi.machine->Primitives)
- return FALSE;
- memset(draw->gs.tgsi.machine->Primitives, 0,
- MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector));
+ for (unsigned i = 0; i < TGSI_MAX_VERTEX_STREAMS; i++) {
+ draw->gs.tgsi.machine->Primitives[i] = align_malloc(
+ MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector), 16);
+ draw->gs.tgsi.machine->PrimitiveOffsets[i] = align_malloc(
+ MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector), 16);
+ if (!draw->gs.tgsi.machine->Primitives[i] || !draw->gs.tgsi.machine->PrimitiveOffsets[i])
+ return FALSE;
+ memset(draw->gs.tgsi.machine->Primitives[i], 0,
+ MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector));
+ memset(draw->gs.tgsi.machine->PrimitiveOffsets[i], 0,
+ MAX_PRIMITIVES * sizeof(struct tgsi_exec_vector));
+ }
}
return TRUE;
void draw_gs_destroy( struct draw_context *draw )
{
+ int i;
if (draw->gs.tgsi.machine) {
- align_free(draw->gs.tgsi.machine->Primitives);
+ for (i = 0; i < TGSI_MAX_VERTEX_STREAMS; i++) {
+ align_free(draw->gs.tgsi.machine->Primitives[i]);
+ align_free(draw->gs.tgsi.machine->PrimitiveOffsets[i]);
+ }
tgsi_exec_machine_destroy(draw->gs.tgsi.machine);
}
}
draw_create_geometry_shader(struct draw_context *draw,
const struct pipe_shader_state *state)
{
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
boolean use_llvm = draw->llvm != NULL;
struct llvm_geometry_shader *llvm_gs = NULL;
#endif
struct draw_geometry_shader *gs;
unsigned i;
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
if (use_llvm) {
llvm_gs = CALLOC_STRUCT(llvm_geometry_shader);
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);
+ if (state->type == PIPE_SHADER_IR_TGSI) {
+ gs->state.tokens = tgsi_dup_tokens(state->tokens);
+ if (!gs->state.tokens) {
+ FREE(gs);
+ return NULL;
+ }
+
+ tgsi_scan_shader(state->tokens, &gs->info);
+ } else
+ nir_tgsi_scan_shader(state->ir.nir, &gs->info, true);
/* setup the defaults */
gs->max_out_prims = 0;
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
if (use_llvm) {
/* TODO: change the input array to handle the following
vector length, instead of the currently hardcoded
gs->machine = draw->gs.tgsi.machine;
-#ifdef HAVE_LLVM
+ gs->num_vertex_streams = 1;
+ for (i = 0; i < gs->state.stream_output.num_outputs; i++) {
+ if (gs->state.stream_output.output[i].stream >= gs->num_vertex_streams)
+ gs->num_vertex_streams = gs->state.stream_output.output[i].stream + 1;
+ }
+
+#ifdef LLVM_AVAILABLE
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->llvm_emitted_primitives = align_malloc(vector_size * gs->num_vertex_streams, vector_size);
+ gs->llvm_emitted_vertices = align_malloc(vector_size * gs->num_vertex_streams, vector_size);
+ gs->llvm_prim_ids = align_calloc(vector_size, vector_size);
gs->fetch_outputs = llvm_fetch_gs_outputs;
gs->fetch_inputs = llvm_fetch_gs_input;
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));
+ gs->info.file_max[TGSI_FILE_SAMPLER_VIEW]+1),
+ gs->info.file_max[TGSI_FILE_IMAGE]+1);
} else
#endif
{
void draw_delete_geometry_shader(struct draw_context *draw,
struct draw_geometry_shader *dgs)
{
+ int i;
if (!dgs) {
return;
}
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
if (draw->llvm) {
struct llvm_geometry_shader *shader = llvm_geometry_shader(dgs);
struct draw_gs_llvm_variant_list_item *li;
if (dgs->llvm_prim_lengths) {
unsigned i;
- for (i = 0; i < dgs->max_out_prims; ++i) {
+ for (i = 0; i < dgs->num_vertex_streams * dgs->max_out_prims; ++i) {
align_free(dgs->llvm_prim_lengths[i]);
}
FREE(dgs->llvm_prim_lengths);
}
#endif
- FREE(dgs->primitive_lengths);
+ for (i = 0; i < TGSI_MAX_VERTEX_STREAMS; i++)
+ FREE(dgs->stream[i].primitive_lengths);
+
+ if (dgs->state.ir.nir)
+ ralloc_free(dgs->state.ir.nir);
FREE((void*) dgs->state.tokens);
FREE(dgs);
}
-#ifdef HAVE_LLVM
+#ifdef LLVM_AVAILABLE
void draw_gs_set_current_variant(struct draw_geometry_shader *shader,
struct draw_gs_llvm_variant *variant)
{