#include "util/u_memory.h"
#include "util/u_prim.h"
+#include "util/disk_cache.h"
+#include "util/mesa-sha1.h"
+
/* SHADER_CACHE */
/**
*/
static bool si_shader_cache_insert_shader(struct si_screen *sscreen,
void *tgsi_binary,
- struct si_shader *shader)
+ struct si_shader *shader,
+ bool insert_into_disk_cache)
{
void *hw_binary;
struct hash_entry *entry;
+ uint8_t key[CACHE_KEY_SIZE];
entry = _mesa_hash_table_search(sscreen->shader_cache, tgsi_binary);
if (entry)
return false;
}
+ if (sscreen->b.disk_shader_cache && insert_into_disk_cache) {
+ _mesa_sha1_compute(tgsi_binary, *((uint32_t *)tgsi_binary), key);
+ disk_cache_put(sscreen->b.disk_shader_cache, key, hw_binary,
+ *((uint32_t *) hw_binary));
+ }
+
return true;
}
{
struct hash_entry *entry =
_mesa_hash_table_search(sscreen->shader_cache, tgsi_binary);
- if (!entry)
- return false;
+ if (!entry) {
+ if (sscreen->b.disk_shader_cache) {
+ unsigned char sha1[CACHE_KEY_SIZE];
+ size_t tg_size = *((uint32_t *) tgsi_binary);
+
+ _mesa_sha1_compute(tgsi_binary, tg_size, sha1);
+
+ size_t binary_size;
+ uint8_t *buffer =
+ disk_cache_get(sscreen->b.disk_shader_cache,
+ sha1, &binary_size);
+ if (!buffer)
+ return false;
- if (!si_load_shader_binary(shader, entry->data))
- return false;
+ if (binary_size < sizeof(uint32_t) ||
+ *((uint32_t*)buffer) != binary_size) {
+ /* Something has gone wrong discard the item
+ * from the cache and rebuild/link from
+ * source.
+ */
+ assert(!"Invalid radeonsi shader disk cache "
+ "item!");
+
+ disk_cache_remove(sscreen->b.disk_shader_cache,
+ sha1);
+ free(buffer);
+
+ return false;
+ }
+
+ if (!si_load_shader_binary(shader, buffer)) {
+ free(buffer);
+ return false;
+ }
+ free(buffer);
+ if (!si_shader_cache_insert_shader(sscreen, tgsi_binary,
+ shader, false))
+ FREE(tgsi_binary);
+ } else {
+ return false;
+ }
+ } else {
+ if (si_load_shader_binary(shader, entry->data))
+ FREE(tgsi_binary);
+ else
+ return false;
+ }
p_atomic_inc(&sscreen->b.num_shader_cache_hits);
return true;
}
bool si_init_shader_cache(struct si_screen *sscreen)
{
- pipe_mutex_init(sscreen->shader_cache_mutex);
+ (void) mtx_init(&sscreen->shader_cache_mutex, mtx_plain);
sscreen->shader_cache =
_mesa_hash_table_create(NULL,
si_shader_cache_key_hash,
si_shader_cache_key_equals);
+
return sscreen->shader_cache != NULL;
}
if (sscreen->shader_cache)
_mesa_hash_table_destroy(sscreen->shader_cache,
si_destroy_shader_cache_entry);
- pipe_mutex_destroy(sscreen->shader_cache_mutex);
+ mtx_destroy(&sscreen->shader_cache_mutex);
}
/* SHADER STATES */
S_028B6C_DISTRIBUTION_MODE(distribution_mode));
}
+/* Polaris needs different VTX_REUSE_DEPTH settings depending on
+ * whether the "fractional odd" tessellation spacing is used.
+ *
+ * Possible VGT configurations and which state should set the register:
+ *
+ * Reg set in | VGT shader configuration | Value
+ * ------------------------------------------------------
+ * VS as VS | VS | 30
+ * VS as ES | ES -> GS -> VS | 30
+ * TES as VS | LS -> HS -> VS | 14 or 30
+ * TES as ES | LS -> HS -> ES -> GS -> VS | 14 or 30
+ */
+static void polaris_set_vgt_vertex_reuse(struct si_screen *sscreen,
+ struct si_shader *shader,
+ struct si_pm4_state *pm4)
+{
+ unsigned type = shader->selector->type;
+
+ if (sscreen->b.family < CHIP_POLARIS10)
+ return;
+
+ /* VS as VS, or VS as ES: */
+ if ((type == PIPE_SHADER_VERTEX &&
+ !shader->key.as_ls &&
+ !shader->is_gs_copy_shader) ||
+ /* TES as VS, or TES as ES: */
+ type == PIPE_SHADER_TESS_EVAL) {
+ unsigned vtx_reuse_depth = 30;
+
+ if (type == PIPE_SHADER_TESS_EVAL &&
+ shader->selector->info.properties[TGSI_PROPERTY_TES_SPACING] ==
+ PIPE_TESS_SPACING_FRACTIONAL_ODD)
+ vtx_reuse_depth = 14;
+
+ si_pm4_set_reg(pm4, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL,
+ vtx_reuse_depth);
+ }
+}
+
static struct si_pm4_state *si_get_shader_pm4_state(struct si_shader *shader)
{
if (shader->pm4)
if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader, pm4);
+
+ polaris_set_vgt_vertex_reuse(sscreen, shader, pm4);
}
/**
if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader, pm4);
+
+ polaris_set_vgt_vertex_reuse(sscreen, shader, pm4);
}
static unsigned si_get_ps_num_interp(struct si_shader *ps)
key->part.vs.prolog.instance_divisors[i] =
sctx->vertex_elements->elements[i].instance_divisor;
- key->mono.vs.fix_fetch =
- sctx->vertex_elements->fix_fetch &
- u_bit_consecutive64(0, 4 * count);
+ memcpy(key->mono.vs.fix_fetch,
+ sctx->vertex_elements->fix_fetch, count);
}
if (sctx->tes_shader.cso)
key->as_ls = 1;
case PIPE_SHADER_TESS_CTRL:
key->part.tcs.epilog.prim_mode =
sctx->tes_shader.cso->info.properties[TGSI_PROPERTY_TES_PRIM_MODE];
+ key->part.tcs.epilog.tes_reads_tess_factors =
+ sctx->tes_shader.cso->info.reads_tess_factors;
if (sel == sctx->fixed_func_tcs_shader.cso)
key->mono.tcs.inputs_to_copy = sctx->vs_shader.cso->outputs_written;
* to the range supported by the type if a channel has less
* than 16 bits and the export format is 16_ABGR.
*/
- if (sctx->b.chip_class <= CIK && sctx->b.family != CHIP_HAWAII)
+ if (sctx->b.chip_class <= CIK && sctx->b.family != CHIP_HAWAII) {
key->part.ps.epilog.color_is_int8 = sctx->framebuffer.color_is_int8;
+ key->part.ps.epilog.color_is_int10 = sctx->framebuffer.color_is_int10;
+ }
/* Disable unwritten outputs (if WRITE_ALL_CBUFS isn't enabled). */
if (!key->part.ps.epilog.last_cbuf) {
key->part.ps.epilog.spi_shader_col_format &= sel->colors_written_4bit;
key->part.ps.epilog.color_is_int8 &= sel->info.colors_written;
+ key->part.ps.epilog.color_is_int10 &= sel->info.colors_written;
}
if (rs) {
* in a compiler thread.
*/
if (thread_index < 0)
- util_queue_job_wait(&sel->ready);
+ util_queue_fence_wait(&sel->ready);
- pipe_mutex_lock(sel->mutex);
+ mtx_lock(&sel->mutex);
/* Find the shader variant. */
for (iter = sel->first_variant; iter; iter = iter->next_variant) {
shader->key = *key;
shader->compiler_ctx_state = *compiler_state;
+ /* Compile the main shader part if it doesn't exist. This can happen
+ * if the initial guess was wrong. */
+ struct si_shader **mainp = si_get_main_shader_part(sel, key);
+ bool is_pure_monolithic =
+ sscreen->use_monolithic_shaders ||
+ memcmp(&key->mono, &zeroed.mono, sizeof(key->mono)) != 0;
+
+ if (!*mainp && !is_pure_monolithic) {
+ struct si_shader *main_part = CALLOC_STRUCT(si_shader);
+
+ if (!main_part) {
+ FREE(shader);
+ pipe_mutex_unlock(sel->mutex);
+ return -ENOMEM; /* skip the draw call */
+ }
+
+ main_part->selector = sel;
+ main_part->key.as_es = key->as_es;
+ main_part->key.as_ls = key->as_ls;
+
+ if (si_compile_tgsi_shader(sscreen, compiler_state->tm,
+ main_part, false,
+ &compiler_state->debug) != 0) {
+ FREE(main_part);
+ FREE(shader);
+ pipe_mutex_unlock(sel->mutex);
+ return -ENOMEM; /* skip the draw call */
+ }
+ *mainp = main_part;
+ }
+
/* Monolithic-only shaders don't make a distinction between optimized
* and unoptimized. */
shader->is_monolithic =
- !sel->main_shader_part ||
- sel->main_shader_part->key.as_ls != key->as_ls ||
- sel->main_shader_part->key.as_es != key->as_es ||
- memcmp(&key->opt, &zeroed.opt, sizeof(key->opt)) != 0 ||
- memcmp(&key->mono, &zeroed.mono, sizeof(key->mono)) != 0;
+ is_pure_monolithic ||
+ memcmp(&key->opt, &zeroed.opt, sizeof(key->opt)) != 0;
shader->is_optimized =
- !sscreen->use_monolithic_shaders &&
+ !is_pure_monolithic &&
memcmp(&key->opt, &zeroed.opt, sizeof(key->opt)) != 0;
if (shader->is_optimized)
util_queue_fence_init(&shader->optimized_ready);
/* If it's an optimized shader, compile it asynchronously. */
if (shader->is_optimized &&
+ !is_pure_monolithic &&
thread_index < 0) {
/* Compile it asynchronously. */
util_queue_add_job(&sscreen->shader_compiler_queue,
tgsi_binary = si_get_tgsi_binary(sel);
/* Try to load the shader from the shader cache. */
- pipe_mutex_lock(sscreen->shader_cache_mutex);
+ mtx_lock(&sscreen->shader_cache_mutex);
if (tgsi_binary &&
si_shader_cache_load_shader(sscreen, tgsi_binary, shader)) {
- FREE(tgsi_binary);
pipe_mutex_unlock(sscreen->shader_cache_mutex);
} else {
pipe_mutex_unlock(sscreen->shader_cache_mutex);
}
if (tgsi_binary) {
- pipe_mutex_lock(sscreen->shader_cache_mutex);
- if (!si_shader_cache_insert_shader(sscreen, tgsi_binary, shader))
+ mtx_lock(&sscreen->shader_cache_mutex);
+ if (!si_shader_cache_insert_shader(sscreen, tgsi_binary, shader, true))
FREE(tgsi_binary);
pipe_mutex_unlock(sscreen->shader_cache_mutex);
}
}
- sel->main_shader_part = shader;
+ *si_get_main_shader_part(sel, &shader->key) = shader;
/* Unset "outputs_written" flags for outputs converted to
* DEFAULT_VAL, so that later inter-shader optimizations don't
sel->db_shader_control |= S_02880C_Z_ORDER(V_02880C_EARLY_Z_THEN_LATE_Z);
}
- pipe_mutex_init(sel->mutex);
+ (void) mtx_init(&sel->mutex, mtx_plain);
util_queue_fence_init(&sel->ready);
if ((sctx->b.debug.debug_message && !sctx->b.debug.async) ||
sctx->gs_shader.cso = sel;
sctx->gs_shader.current = sel ? sel->first_variant : NULL;
+ sctx->ia_multi_vgt_param_key.u.uses_gs = sel != NULL;
sctx->do_update_shaders = true;
si_mark_atom_dirty(sctx, &sctx->clip_regs);
sctx->last_rast_prim = -1; /* reset this so that it gets updated */
r600_update_vs_writes_viewport_index(&sctx->b, si_get_vs_info(sctx));
}
+static void si_update_tcs_tes_uses_prim_id(struct si_context *sctx)
+{
+ sctx->ia_multi_vgt_param_key.u.tcs_tes_uses_prim_id =
+ (sctx->tes_shader.cso &&
+ sctx->tes_shader.cso->info.uses_primid) ||
+ (sctx->tcs_shader.cso &&
+ sctx->tcs_shader.cso->info.uses_primid);
+}
+
static void si_bind_tcs_shader(struct pipe_context *ctx, void *state)
{
struct si_context *sctx = (struct si_context *)ctx;
sctx->tcs_shader.cso = sel;
sctx->tcs_shader.current = sel ? sel->first_variant : NULL;
+ si_update_tcs_tes_uses_prim_id(sctx);
sctx->do_update_shaders = true;
if (enable_changed)
sctx->tes_shader.cso = sel;
sctx->tes_shader.current = sel ? sel->first_variant : NULL;
+ sctx->ia_multi_vgt_param_key.u.uses_tess = sel != NULL;
+ si_update_tcs_tes_uses_prim_id(sctx);
sctx->do_update_shaders = true;
si_mark_atom_dirty(sctx, &sctx->clip_regs);
sctx->last_rast_prim = -1; /* reset this so that it gets updated */
static void si_delete_shader(struct si_context *sctx, struct si_shader *shader)
{
if (shader->is_optimized) {
- util_queue_job_wait(&shader->optimized_ready);
+ util_queue_fence_wait(&shader->optimized_ready);
util_queue_fence_destroy(&shader->optimized_ready);
}
[PIPE_SHADER_FRAGMENT] = &sctx->ps_shader,
};
- util_queue_job_wait(&sel->ready);
+ util_queue_fence_wait(&sel->ready);
if (current_shader[sel->type]->cso == sel) {
current_shader[sel->type]->cso = NULL;
if (sel->main_shader_part)
si_delete_shader(sctx, sel->main_shader_part);
+ if (sel->main_shader_part_ls)
+ si_delete_shader(sctx, sel->main_shader_part_ls);
+ if (sel->main_shader_part_es)
+ si_delete_shader(sctx, sel->main_shader_part_es);
if (sel->gs_copy_shader)
si_delete_shader(sctx, sel->gs_copy_shader);
util_queue_fence_destroy(&sel->ready);
- pipe_mutex_destroy(sel->mutex);
+ mtx_destroy(&sel->mutex);
free(sel->tokens);
free(sel);
}
if (update_esgs) {
pipe_resource_reference(&sctx->esgs_ring, NULL);
- sctx->esgs_ring = pipe_buffer_create(sctx->b.b.screen, 0,
- PIPE_USAGE_DEFAULT,
- esgs_ring_size);
+ sctx->esgs_ring =
+ r600_aligned_buffer_create(sctx->b.b.screen,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
+ PIPE_USAGE_DEFAULT,
+ esgs_ring_size, alignment);
if (!sctx->esgs_ring)
return false;
}
if (update_gsvs) {
pipe_resource_reference(&sctx->gsvs_ring, NULL);
- sctx->gsvs_ring = pipe_buffer_create(sctx->b.b.screen, 0,
- PIPE_USAGE_DEFAULT,
- gsvs_ring_size);
+ sctx->gsvs_ring =
+ r600_aligned_buffer_create(sctx->b.b.screen,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
+ PIPE_USAGE_DEFAULT,
+ gsvs_ring_size, alignment);
if (!sctx->gsvs_ring)
return false;
}
r600_resource_reference(&sctx->scratch_buffer, NULL);
sctx->scratch_buffer = (struct r600_resource*)
- pipe_buffer_create(&sctx->screen->b.b, 0,
- PIPE_USAGE_DEFAULT, scratch_needed_size);
+ r600_aligned_buffer_create(&sctx->screen->b.b,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
+ PIPE_USAGE_DEFAULT,
+ scratch_needed_size, 256);
if (!sctx->scratch_buffer)
return false;
- sctx->emit_scratch_reloc = true;
+
+ si_mark_atom_dirty(sctx, &sctx->scratch_state);
+ r600_context_add_resource_size(&sctx->b.b,
+ &sctx->scratch_buffer->b.b);
}
/* Update the shaders, so they are using the latest scratch. The
S_0286E8_WAVESIZE(scratch_bytes_per_wave >> 10);
if (spi_tmpring_size != sctx->spi_tmpring_size) {
sctx->spi_tmpring_size = spi_tmpring_size;
- sctx->emit_scratch_reloc = true;
+ si_mark_atom_dirty(sctx, &sctx->scratch_state);
}
return true;
}
static void si_init_tess_factor_ring(struct si_context *sctx)
{
- bool double_offchip_buffers = sctx->b.chip_class >= CIK;
+ bool double_offchip_buffers = sctx->b.chip_class >= CIK &&
+ sctx->b.family != CHIP_CARRIZO &&
+ sctx->b.family != CHIP_STONEY;
unsigned max_offchip_buffers_per_se = double_offchip_buffers ? 128 : 64;
unsigned max_offchip_buffers = max_offchip_buffers_per_se *
sctx->screen->b.info.max_se;
}
assert(!sctx->tf_ring);
- sctx->tf_ring = pipe_buffer_create(sctx->b.b.screen, 0,
- PIPE_USAGE_DEFAULT,
- 32768 * sctx->screen->b.info.max_se);
+ sctx->tf_ring = r600_aligned_buffer_create(sctx->b.b.screen,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
+ PIPE_USAGE_DEFAULT,
+ 32768 * sctx->screen->b.info.max_se,
+ 256);
if (!sctx->tf_ring)
return;
assert(((sctx->tf_ring->width0 / 4) & C_030938_SIZE) == 0);
- sctx->tess_offchip_ring = pipe_buffer_create(sctx->b.b.screen, 0,
- PIPE_USAGE_DEFAULT,
- max_offchip_buffers *
- sctx->screen->tess_offchip_block_dw_size * 4);
+ sctx->tess_offchip_ring =
+ r600_aligned_buffer_create(sctx->b.b.screen,
+ R600_RESOURCE_FLAG_UNMAPPABLE,
+ PIPE_USAGE_DEFAULT,
+ max_offchip_buffers *
+ sctx->screen->tess_offchip_block_dw_size * 4,
+ 256);
if (!sctx->tess_offchip_ring)
return;
return false;
}
+ if (sctx->b.chip_class >= CIK)
+ si_mark_atom_dirty(sctx, &sctx->prefetch_L2);
+
sctx->do_update_shaders = false;
return true;
}
+static void si_emit_scratch_state(struct si_context *sctx,
+ struct r600_atom *atom)
+{
+ struct radeon_winsys_cs *cs = sctx->b.gfx.cs;
+
+ radeon_set_context_reg(cs, R_0286E8_SPI_TMPRING_SIZE,
+ sctx->spi_tmpring_size);
+
+ if (sctx->scratch_buffer) {
+ radeon_add_to_buffer_list(&sctx->b, &sctx->b.gfx,
+ sctx->scratch_buffer, RADEON_USAGE_READWRITE,
+ RADEON_PRIO_SCRATCH_BUFFER);
+ }
+}
+
void si_init_shader_functions(struct si_context *sctx)
{
si_init_atom(sctx, &sctx->spi_map, &sctx->atoms.s.spi_map, si_emit_spi_map);
+ si_init_atom(sctx, &sctx->scratch_state, &sctx->atoms.s.scratch_state,
+ si_emit_scratch_state);
sctx->b.b.create_vs_state = si_create_shader_selector;
sctx->b.b.create_tcs_state = si_create_shader_selector;