#include "brw_vec4_gs_visitor.h"
#include "brw_state.h"
#include "brw_ff_gs.h"
+#include "brw_nir.h"
+static void
+assign_gs_binding_table_offsets(const struct brw_device_info *devinfo,
+ const struct gl_shader_program *shader_prog,
+ const struct gl_program *prog,
+ struct brw_gs_prog_data *prog_data)
+{
+ /* In gen6 we reserve the first BRW_MAX_SOL_BINDINGS entries for transform
+ * feedback surfaces.
+ */
+ uint32_t reserved = devinfo->gen == 6 ? BRW_MAX_SOL_BINDINGS : 0;
-static bool
-do_gs_prog(struct brw_context *brw,
- struct gl_shader_program *prog,
- struct brw_geometry_program *gp,
- struct brw_gs_prog_key *key)
+ brw_assign_common_binding_table_offsets(MESA_SHADER_GEOMETRY, devinfo,
+ shader_prog, prog,
+ &prog_data->base.base,
+ reserved);
+}
+
+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;
c.prog_data.invocations = gp->program.Invocations;
+ assign_gs_binding_table_offsets(brw->intelScreen->devinfo, prog,
+ &gp->program.Base, &c.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.
* 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;
+ int param_count = gp->program.Base.nir->num_uniforms * 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);
- /* Setting nr_params here NOT to the size of the param and pull_param
- * arrays, but to the number of uniform components vec4_visitor
- * needs. vec4_visitor::setup_uniforms() will set it back to a proper value.
- */
- c.prog_data.base.base.nr_params = ALIGN(param_count, 4) / 4 + gs->num_samplers;
+ c.prog_data.base.base.image_param =
+ rzalloc_array(NULL, struct brw_image_param, gs->NumImages);
+ c.prog_data.base.base.nr_params = param_count;
+ c.prog_data.base.base.nr_image_params = gs->NumImages;
+
+ brw_nir_setup_glsl_uniforms(gp->program.Base.nir, prog, &gp->program.Base,
+ &c.prog_data.base.base, false);
+
+ if (brw->gen >= 8) {
+ c.prog_data.static_vertex_count = !gp->program.Base.nir ? -1 :
+ nir_gs_count_vertices(gp->program.Base.nir);
+ }
if (brw->gen >= 7) {
if (gp->program.OutputType == GL_POINTS) {
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, &c.prog_data.base.vue_map, outputs_written);
+ brw_compute_vue_map(brw->intelScreen->devinfo,
+ &c.prog_data.base.vue_map, outputs_written,
+ prog ? prog->SeparateShader : false);
/* Compute the output vertex size.
*
c.prog_data.output_topology =
get_hw_prim_for_gl_prim(gp->program.OutputType);
- brw_compute_vue_map(brw, &c.input_vue_map, c.key.input_varyings);
+ /* The GLSL linker will have already matched up GS inputs and the outputs
+ * of prior stages. The driver does extend VS outputs in some cases, but
+ * only for legacy OpenGL or Gen4-5 hardware, neither of which offer
+ * geometry shader support. So we can safely ignore that.
+ *
+ * For SSO pipelines, we use a fixed VUE map layout based on variable
+ * locations, so we can rely on rendezvous-by-location making this work.
+ *
+ * However, we need to ignore VARYING_SLOT_PRIMITIVE_ID, as it's not
+ * written by previous stages and shows up via payload magic.
+ */
+ GLbitfield64 inputs_read =
+ gp->program.Base.InputsRead & ~VARYING_BIT_PRIMITIVE_ID;
+ brw_compute_vue_map(brw->intelScreen->devinfo,
+ &c.input_vue_map, inputs_read,
+ prog->SeparateShader);
/* 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;
+ if (unlikely(INTEL_DEBUG & DEBUG_GS))
+ brw_dump_ir("geometry", prog, gs, NULL);
+
void *mem_ctx = ralloc_context(NULL);
unsigned program_size;
const unsigned *program =
}
/* 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);
-
+ if (c.prog_data.base.base.total_scratch) {
brw_get_scratch_bo(brw, &stage_state->scratch_bo,
c.prog_data.base.base.total_scratch *
brw->max_gs_threads);
return true;
}
+static bool
+brw_gs_state_dirty(struct brw_context *brw)
+{
+ return brw_state_dirty(brw,
+ _NEW_TEXTURE,
+ BRW_NEW_GEOMETRY_PROGRAM |
+ BRW_NEW_TRANSFORM_FEEDBACK);
+}
static void
+brw_gs_populate_key(struct brw_context *brw,
+ struct brw_gs_prog_key *key)
+{
+ struct gl_context *ctx = &brw->ctx;
+ struct brw_stage_state *stage_state = &brw->gs.base;
+ struct brw_geometry_program *gp =
+ (struct brw_geometry_program *) brw->geometry_program;
+ struct gl_program *prog = &gp->program.Base;
+
+ memset(key, 0, sizeof(*key));
+
+ key->program_string_id = gp->id;
+
+ /* _NEW_TEXTURE */
+ brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count,
+ &key->tex);
+}
+
+void
brw_upload_gs_prog(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
+ struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
struct brw_stage_state *stage_state = &brw->gs.base;
struct brw_gs_prog_key key;
/* BRW_NEW_GEOMETRY_PROGRAM */
struct brw_geometry_program *gp =
(struct brw_geometry_program *) brw->geometry_program;
+ if (!brw_gs_state_dirty(brw))
+ return;
+
if (gp == NULL) {
/* No geometry shader. Vertex data just passes straight through. */
- if (brw->state.dirty.brw & BRW_NEW_VUE_MAP_VS) {
- brw->vue_map_geom_out = brw->vue_map_vs;
- brw->state.dirty.brw |= BRW_NEW_VUE_MAP_GEOM_OUT;
- }
-
if (brw->gen == 6 &&
- (brw->state.dirty.brw & BRW_NEW_TRANSFORM_FEEDBACK)) {
+ (brw->ctx.NewDriverState & BRW_NEW_TRANSFORM_FEEDBACK)) {
gen6_brw_upload_ff_gs_prog(brw);
return;
}
return;
}
- struct gl_program *prog = &gp->program.Base;
-
- memset(&key, 0, sizeof(key));
-
- key.base.program_string_id = gp->id;
- brw_setup_vec4_key_clip_info(brw, &key.base,
- gp->program.Base.UsesClipDistanceOut);
-
- /* _NEW_TEXTURE */
- brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count,
- &key.base.tex);
-
- /* BRW_NEW_VUE_MAP_VS */
- key.input_varyings = brw->vue_map_vs.slots_valid;
+ brw_gs_populate_key(brw, &key);
if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
&key, sizeof(key),
&stage_state->prog_offset, &brw->gs.prog_data)) {
- bool success =
- do_gs_prog(brw, ctx->_Shader->CurrentProgram[MESA_SHADER_GEOMETRY], gp,
- &key);
+ bool success = brw_codegen_gs_prog(brw, current[MESA_SHADER_GEOMETRY],
+ gp, &key);
assert(success);
(void)success;
}
brw->gs.base.prog_data = &brw->gs.prog_data->base.base;
-
- if (memcmp(&brw->vs.prog_data->base.vue_map, &brw->vue_map_geom_out,
- sizeof(brw->vue_map_geom_out)) != 0) {
- brw->vue_map_geom_out = brw->gs.prog_data->base.vue_map;
- brw->state.dirty.brw |= BRW_NEW_VUE_MAP_GEOM_OUT;
- }
}
-
-const struct brw_tracked_state brw_gs_prog = {
- .dirty = {
- .mesa = _NEW_TEXTURE,
- .brw = BRW_NEW_GEOMETRY_PROGRAM |
- BRW_NEW_TRANSFORM_FEEDBACK |
- BRW_NEW_VUE_MAP_VS,
- },
- .emit = brw_upload_gs_prog
-};
-
-
bool
brw_gs_precompile(struct gl_context *ctx,
struct gl_shader_program *shader_prog,
memset(&key, 0, sizeof(key));
- brw_vec4_setup_prog_key_for_precompile(ctx, &key.base, bgp->id, &gp->Base);
-
- /* Assume that the set of varyings coming in from the vertex shader exactly
- * matches what the geometry shader requires.
- */
- key.input_varyings = gp->Base.InputsRead;
+ brw_setup_tex_for_precompile(brw, &key.tex, prog);
+ key.program_string_id = bgp->id;
- success = do_gs_prog(brw, shader_prog, bgp, &key);
+ success = brw_codegen_gs_prog(brw, shader_prog, bgp, &key);
brw->gs.base.prog_offset = old_prog_offset;
brw->gs.prog_data = old_prog_data;
return success;
}
-
-
-bool
-brw_gs_prog_data_compare(const void *in_a, const void *in_b)
-{
- const struct brw_gs_prog_data *a = in_a;
- const struct brw_gs_prog_data *b = in_b;
-
- /* Compare the base structure. */
- if (!brw_stage_prog_data_compare(&a->base.base, &b->base.base))
- return false;
-
- /* Compare the rest of the struct. */
- const unsigned offset = sizeof(struct brw_stage_prog_data);
- if (memcmp(((char *) a) + offset, ((char *) b) + offset,
- sizeof(struct brw_gs_prog_data) - offset)) {
- return false;
- }
-
- return true;
-}
projects / mesa.git / blobdiff