return false;
return shader->key.as_ls || shader->key.as_es ||
- shader->selector->type == PIPE_SHADER_TESS_CTRL ||
- shader->selector->type == PIPE_SHADER_GEOMETRY;
+ shader->selector->info.stage == MESA_SHADER_TESS_CTRL ||
+ shader->selector->info.stage == MESA_SHADER_GEOMETRY;
}
/** Whether the shader runs on a merged HW stage (LSHS or ESGS) */
unsigned si_get_max_workgroup_size(const struct si_shader *shader)
{
- switch (shader->selector->type) {
- case PIPE_SHADER_VERTEX:
- case PIPE_SHADER_TESS_EVAL:
+ switch (shader->selector->info.stage) {
+ case MESA_SHADER_VERTEX:
+ case MESA_SHADER_TESS_EVAL:
return shader->key.as_ngg ? 128 : 0;
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
/* Return this so that LLVM doesn't remove s_barrier
* instructions on chips where we use s_barrier. */
return shader->selector->screen->info.chip_class >= GFX7 ? 128 : 0;
- case PIPE_SHADER_GEOMETRY:
+ case MESA_SHADER_GEOMETRY:
return shader->selector->screen->info.chip_class >= GFX9 ? 128 : 0;
- case PIPE_SHADER_COMPUTE:
+ case MESA_SHADER_COMPUTE:
break; /* see below */
default:
unsigned num_lds_symbols = 0;
if (sel && screen->info.chip_class >= GFX9 && !shader->is_gs_copy_shader &&
- (sel->type == PIPE_SHADER_GEOMETRY || shader->key.as_ngg)) {
+ (sel->info.stage == MESA_SHADER_GEOMETRY || shader->key.as_ngg)) {
/* We add this symbol even on LLVM <= 8 to ensure that
* shader->config.lds_size is set correctly below.
*/
sym->align = 64 * 1024;
}
- if (shader->key.as_ngg && sel->type == PIPE_SHADER_GEOMETRY) {
+ if (shader->key.as_ngg && sel->info.stage == MESA_SHADER_GEOMETRY) {
struct ac_rtld_symbol *sym = &lds_symbols[num_lds_symbols++];
sym->name = "ngg_emit";
sym->size = shader->ngg.ngg_emit_size * 4;
{
.halt_at_entry = screen->options.halt_shaders,
},
- .shader_type = tgsi_processor_to_shader_stage(sel->type),
+ .shader_type = sel->info.stage,
.wave_size = si_get_shader_wave_size(shader),
.num_parts = num_parts,
.elf_ptrs = part_elfs,
max_simd_waves = sscreen->info.max_wave64_per_simd;
/* Compute LDS usage for PS. */
- switch (shader->selector->type) {
- case PIPE_SHADER_FRAGMENT:
+ switch (shader->selector->info.stage) {
+ case MESA_SHADER_FRAGMENT:
/* The minimum usage per wave is (num_inputs * 48). The maximum
* usage is (num_inputs * 48 * 16).
* We can get anything in between and it varies between waves.
*/
lds_per_wave = conf->lds_size * lds_increment + align(num_inputs * 48, lds_increment);
break;
- case PIPE_SHADER_COMPUTE:
+ case MESA_SHADER_COMPUTE:
if (shader->selector) {
unsigned max_workgroup_size = si_get_max_workgroup_size(shader);
lds_per_wave = (conf->lds_size * lds_increment) /
const struct ac_shader_config *conf = &shader->config;
if (!check_debug_option || si_can_dump_shader(sscreen, shader->selector->type)) {
- if (shader->selector->type == PIPE_SHADER_FRAGMENT) {
+ if (shader->selector->info.stage == MESA_SHADER_FRAGMENT) {
fprintf(file,
"*** SHADER CONFIG ***\n"
"SPI_PS_INPUT_ADDR = 0x%04x\n"
const char *si_get_shader_name(const struct si_shader *shader)
{
- switch (shader->selector->type) {
- case PIPE_SHADER_VERTEX:
+ switch (shader->selector->info.stage) {
+ case MESA_SHADER_VERTEX:
if (shader->key.as_es)
return "Vertex Shader as ES";
else if (shader->key.as_ls)
return "Vertex Shader as ESGS";
else
return "Vertex Shader as VS";
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
return "Tessellation Control Shader";
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_TESS_EVAL:
if (shader->key.as_es)
return "Tessellation Evaluation Shader as ES";
else if (shader->key.as_ngg)
return "Tessellation Evaluation Shader as ESGS";
else
return "Tessellation Evaluation Shader as VS";
- case PIPE_SHADER_GEOMETRY:
+ case MESA_SHADER_GEOMETRY:
if (shader->is_gs_copy_shader)
return "GS Copy Shader as VS";
else
return "Geometry Shader";
- case PIPE_SHADER_FRAGMENT:
+ case MESA_SHADER_FRAGMENT:
return "Pixel Shader";
- case PIPE_SHADER_COMPUTE:
+ case MESA_SHADER_COMPUTE:
return "Compute Shader";
default:
return "Unknown Shader";
static void si_dump_shader_key(const struct si_shader *shader, FILE *f)
{
const struct si_shader_key *key = &shader->key;
- enum pipe_shader_type shader_type = shader->selector->type;
+ gl_shader_stage stage = shader->selector->info.stage;
fprintf(f, "SHADER KEY\n");
- switch (shader_type) {
- case PIPE_SHADER_VERTEX:
+ switch (stage) {
+ case MESA_SHADER_VERTEX:
si_dump_shader_key_vs(key, &key->part.vs.prolog, "part.vs.prolog", f);
fprintf(f, " as_es = %u\n", key->as_es);
fprintf(f, " as_ls = %u\n", key->as_ls);
fprintf(f, " opt.cs_halfz_clip_space = %u\n", key->opt.cs_halfz_clip_space);
break;
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
if (shader->selector->screen->info.chip_class >= GFX9) {
si_dump_shader_key_vs(key, &key->part.tcs.ls_prolog, "part.tcs.ls_prolog", f);
}
key->mono.u.ff_tcs_inputs_to_copy);
break;
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_TESS_EVAL:
fprintf(f, " as_es = %u\n", key->as_es);
fprintf(f, " as_ngg = %u\n", key->as_ngg);
fprintf(f, " mono.u.vs_export_prim_id = %u\n", key->mono.u.vs_export_prim_id);
break;
- case PIPE_SHADER_GEOMETRY:
+ case MESA_SHADER_GEOMETRY:
if (shader->is_gs_copy_shader)
break;
if (shader->selector->screen->info.chip_class >= GFX9 &&
- key->part.gs.es->type == PIPE_SHADER_VERTEX) {
+ key->part.gs.es->info.stage == MESA_SHADER_VERTEX) {
si_dump_shader_key_vs(key, &key->part.gs.vs_prolog, "part.gs.vs_prolog", f);
}
fprintf(f, " part.gs.prolog.tri_strip_adj_fix = %u\n",
fprintf(f, " as_ngg = %u\n", key->as_ngg);
break;
- case PIPE_SHADER_COMPUTE:
+ case MESA_SHADER_COMPUTE:
break;
- case PIPE_SHADER_FRAGMENT:
+ case MESA_SHADER_FRAGMENT:
fprintf(f, " part.ps.prolog.color_two_side = %u\n", key->part.ps.prolog.color_two_side);
fprintf(f, " part.ps.prolog.flatshade_colors = %u\n", key->part.ps.prolog.flatshade_colors);
fprintf(f, " part.ps.prolog.poly_stipple = %u\n", key->part.ps.prolog.poly_stipple);
assert(0);
}
- if ((shader_type == PIPE_SHADER_GEOMETRY || shader_type == PIPE_SHADER_TESS_EVAL ||
- shader_type == PIPE_SHADER_VERTEX) &&
+ if ((stage == MESA_SHADER_GEOMETRY || stage == MESA_SHADER_TESS_EVAL ||
+ stage == MESA_SHADER_VERTEX) &&
!key->as_es && !key->as_ls) {
fprintf(f, " opt.kill_outputs = 0x%" PRIx64 "\n", key->opt.kill_outputs);
fprintf(f, " opt.clip_disable = %u\n", key->opt.clip_disable);
- if (shader_type != PIPE_SHADER_GEOMETRY)
+ if (stage != MESA_SHADER_GEOMETRY)
fprintf(f, " opt.ngg_culling = 0x%x\n", key->opt.ngg_culling);
}
}
key->vs_prolog.has_ngg_cull_inputs = !!shader_out->key.opt.ngg_culling;
}
- if (shader_out->selector->type == PIPE_SHADER_TESS_CTRL) {
+ if (shader_out->selector->info.stage == MESA_SHADER_TESS_CTRL) {
key->vs_prolog.as_ls = 1;
key->vs_prolog.num_merged_next_stage_vgprs = 2;
- } else if (shader_out->selector->type == PIPE_SHADER_GEOMETRY) {
+ } else if (shader_out->selector->info.stage == MESA_SHADER_GEOMETRY) {
key->vs_prolog.as_es = 1;
key->vs_prolog.num_merged_next_stage_vgprs = 5;
} else if (shader_out->key.as_ngg) {
/* For a crazy dEQP test containing 2597 memory opcodes, mostly
* buffer stores. */
- return sel->type == PIPE_SHADER_COMPUTE && sel->info.num_memory_instructions > 1000;
+ return sel->info.stage == MESA_SHADER_COMPUTE && sel->info.num_memory_instructions > 1000;
}
static struct nir_shader *get_nir_shader(struct si_shader_selector *sel, bool *free_nir)
es_main = ctx.main_fn;
/* ES prolog */
- if (es->type == PIPE_SHADER_VERTEX &&
+ if (es->info.stage == MESA_SHADER_VERTEX &&
si_vs_needs_prolog(es, &shader->key.part.gs.vs_prolog, &shader->key, false)) {
union si_shader_part_key vs_prolog_key;
si_get_vs_prolog_key(&es->info, shader_es.info.num_input_sgprs, false,
/* Validate SGPR and VGPR usage for compute to detect compiler bugs.
* LLVM 3.9svn has this bug.
*/
- if (sel->type == PIPE_SHADER_COMPUTE) {
+ if (sel->info.stage == MESA_SHADER_COMPUTE) {
unsigned wave_size = sscreen->compute_wave_size;
unsigned max_vgprs =
sscreen->info.num_physical_wave64_vgprs_per_simd * (wave_size == 32 ? 2 : 1);
shader->info.num_input_sgprs += 1; /* scratch byte offset */
/* Calculate the number of fragment input VGPRs. */
- if (sel->type == PIPE_SHADER_FRAGMENT) {
+ if (sel->info.stage == MESA_SHADER_FRAGMENT) {
shader->info.num_input_vgprs = ac_get_fs_input_vgpr_cnt(
&shader->config, &shader->info.face_vgpr_index, &shader->info.ancillary_vgpr_index);
}
struct si_shader_context ctx;
si_llvm_context_init(&ctx, sscreen, compiler,
- si_get_wave_size(sscreen, type, shader.key.as_ngg, shader.key.as_es,
+ si_get_wave_size(sscreen, tgsi_processor_to_shader_stage(type),
+ shader.key.as_ngg, shader.key.as_es,
shader.key.opt.ngg_culling & SI_NGG_CULL_GS_FAST_LAUNCH_ALL,
shader.key.opt.vs_as_prim_discard_cs));
ctx.shader = &shader;
{
if (sscreen->info.chip_class >= GFX9) {
struct si_shader *es_main_part;
- enum pipe_shader_type es_type = shader->key.part.gs.es->type;
if (shader->key.as_ngg)
es_main_part = shader->key.part.gs.es->main_shader_part_ngg_es;
else
es_main_part = shader->key.part.gs.es->main_shader_part_es;
- if (es_type == PIPE_SHADER_VERTEX &&
+ if (shader->key.part.gs.es->info.stage == MESA_SHADER_VERTEX &&
!si_get_vs_prolog(sscreen, compiler, shader, debug, es_main_part,
&shader->key.part.gs.vs_prolog))
return false;
shader->config.num_sgprs = MAX2(shader->config.num_sgprs, min_sgprs);
- if (shader->selector->type == PIPE_SHADER_COMPUTE &&
+ if (shader->selector->info.stage == MESA_SHADER_COMPUTE &&
si_get_max_workgroup_size(shader) > sscreen->compute_wave_size) {
si_multiwave_lds_size_workaround(sscreen, &shader->config.lds_size);
}
shader->info.nr_param_exports = mainp->info.nr_param_exports;
/* Select prologs and/or epilogs. */
- switch (sel->type) {
- case PIPE_SHADER_VERTEX:
+ switch (sel->info.stage) {
+ case MESA_SHADER_VERTEX:
if (!si_shader_select_vs_parts(sscreen, compiler, shader, debug))
return false;
break;
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
if (!si_shader_select_tcs_parts(sscreen, compiler, shader, debug))
return false;
break;
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_TESS_EVAL:
break;
- case PIPE_SHADER_GEOMETRY:
+ case MESA_SHADER_GEOMETRY:
if (!si_shader_select_gs_parts(sscreen, compiler, shader, debug))
return false;
break;
- case PIPE_SHADER_FRAGMENT:
+ case MESA_SHADER_FRAGMENT:
if (!si_shader_select_ps_parts(sscreen, compiler, shader, debug))
return false;
fprintf(stderr, "Failed to compute subgroup info\n");
return false;
}
- } else if (sscreen->info.chip_class >= GFX9 && sel->type == PIPE_SHADER_GEOMETRY) {
+ } else if (sscreen->info.chip_class >= GFX9 && sel->info.stage == MESA_SHADER_GEOMETRY) {
gfx9_get_gs_info(shader->previous_stage_sel, sel, &shader->gs_info);
}
#include "util/u_async_debug.h"
#include "util/u_memory.h"
#include "util/u_prim.h"
+#include "tgsi/tgsi_from_mesa.h"
/* SHADER_CACHE */
shader_variant_flags |= 1 << 0;
if (sel->nir)
shader_variant_flags |= 1 << 1;
- if (si_get_wave_size(sel->screen, sel->type, ngg, es, false, false) == 32)
+ if (si_get_wave_size(sel->screen, sel->info.stage, ngg, es, false, false) == 32)
shader_variant_flags |= 1 << 2;
- if (sel->type == PIPE_SHADER_FRAGMENT && sel->info.uses_derivatives && sel->info.uses_kill &&
+ if (sel->info.stage == MESA_SHADER_FRAGMENT && sel->info.uses_derivatives && sel->info.uses_kill &&
sel->screen->debug_flags & DBG(FS_CORRECT_DERIVS_AFTER_KILL))
shader_variant_flags |= 1 << 3;
_mesa_sha1_init(&ctx);
_mesa_sha1_update(&ctx, &shader_variant_flags, 4);
_mesa_sha1_update(&ctx, ir_binary, ir_size);
- if (sel->type == PIPE_SHADER_VERTEX || sel->type == PIPE_SHADER_TESS_EVAL ||
- sel->type == PIPE_SHADER_GEOMETRY)
+ if (sel->info.stage == MESA_SHADER_VERTEX || sel->info.stage == MESA_SHADER_TESS_EVAL ||
+ sel->info.stage == MESA_SHADER_GEOMETRY)
_mesa_sha1_update(&ctx, &sel->so, sizeof(sel->so));
_mesa_sha1_final(&ctx, ir_sha1_cache_key);
static void polaris_set_vgt_vertex_reuse(struct si_screen *sscreen, struct si_shader_selector *sel,
struct si_shader *shader, struct si_pm4_state *pm4)
{
- unsigned type = sel->type;
-
if (sscreen->info.family < CHIP_POLARIS10 || sscreen->info.chip_class >= GFX10)
return;
/* VS as VS, or VS as ES: */
- if ((type == PIPE_SHADER_VERTEX &&
+ if ((sel->info.stage == MESA_SHADER_VERTEX &&
(!shader || (!shader->key.as_ls && !shader->is_gs_copy_shader))) ||
/* TES as VS, or TES as ES: */
- type == PIPE_SHADER_TESS_EVAL) {
+ sel->info.stage == MESA_SHADER_TESS_EVAL) {
unsigned vtx_reuse_depth = 30;
- if (type == PIPE_SHADER_TESS_EVAL &&
+ if (sel->info.stage == MESA_SHADER_TESS_EVAL &&
sel->info.properties[TGSI_PROPERTY_TES_SPACING] == PIPE_TESS_SPACING_FRACTIONAL_ODD)
vtx_reuse_depth = 14;
static unsigned si_get_vs_vgpr_comp_cnt(struct si_screen *sscreen, struct si_shader *shader,
bool legacy_vs_prim_id)
{
- assert(shader->selector->type == PIPE_SHADER_VERTEX ||
- (shader->previous_stage_sel && shader->previous_stage_sel->type == PIPE_SHADER_VERTEX));
+ assert(shader->selector->info.stage == MESA_SHADER_VERTEX ||
+ (shader->previous_stage_sel && shader->previous_stage_sel->info.stage == MESA_SHADER_VERTEX));
/* GFX6-9 LS (VertexID, RelAutoindex, InstanceID / StepRate0(==1), ...).
* GFX6-9 ES,VS (VertexID, InstanceID / StepRate0(==1), VSPrimID, ...)
* GFX10 ES,VS (VertexID, UserVGPR0, UserVGPR1 or VSPrimID, UserVGPR2 or
* InstanceID)
*/
- bool is_ls = shader->selector->type == PIPE_SHADER_TESS_CTRL || shader->key.as_ls;
+ bool is_ls = shader->selector->info.stage == MESA_SHADER_TESS_CTRL || shader->key.as_ls;
if (sscreen->info.chip_class >= GFX10 && shader->info.uses_instanceid)
return 3;
SI_TRACKED_VGT_ESGS_RING_ITEMSIZE,
shader->selector->esgs_itemsize / 4);
- if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
+ if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL)
radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM, SI_TRACKED_VGT_TF_PARAM,
shader->vgt_tf_param);
pm4->atom.emit = si_emit_shader_es;
va = shader->bo->gpu_address;
- if (shader->selector->type == PIPE_SHADER_VERTEX) {
+ if (shader->selector->info.stage == MESA_SHADER_VERTEX) {
vgpr_comp_cnt = si_get_vs_vgpr_comp_cnt(sscreen, shader, false);
num_user_sgprs = si_get_num_vs_user_sgprs(shader, SI_VS_NUM_USER_SGPR);
- } else if (shader->selector->type == PIPE_SHADER_TESS_EVAL) {
+ } else if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL) {
vgpr_comp_cnt = shader->selector->info.uses_primid ? 3 : 2;
num_user_sgprs = SI_TES_NUM_USER_SGPR;
} else
unreachable("invalid shader selector type");
- oc_lds_en = shader->selector->type == PIPE_SHADER_TESS_EVAL ? 1 : 0;
+ oc_lds_en = shader->selector->info.stage == MESA_SHADER_TESS_EVAL ? 1 : 0;
si_pm4_set_reg(pm4, R_00B320_SPI_SHADER_PGM_LO_ES, va >> 8);
si_pm4_set_reg(pm4, R_00B324_SPI_SHADER_PGM_HI_ES, S_00B324_MEM_BASE(va >> 40));
S_00B32C_USER_SGPR(num_user_sgprs) | S_00B32C_OC_LDS_EN(oc_lds_en) |
S_00B32C_SCRATCH_EN(shader->config.scratch_bytes_per_wave > 0));
- if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
+ if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader->selector, pm4);
polaris_set_vgt_vertex_reuse(sscreen, shader->selector, shader, pm4);
SI_TRACKED_VGT_ESGS_RING_ITEMSIZE,
shader->ctx_reg.gs.vgt_esgs_ring_itemsize);
- if (shader->key.part.gs.es->type == PIPE_SHADER_TESS_EVAL)
+ if (shader->key.part.gs.es->info.stage == MESA_SHADER_TESS_EVAL)
radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM, SI_TRACKED_VGT_TF_PARAM,
shader->vgt_tf_param);
if (shader->vgt_vertex_reuse_block_cntl)
if (sscreen->info.chip_class >= GFX9) {
unsigned input_prim = sel->info.properties[TGSI_PROPERTY_GS_INPUT_PRIM];
- unsigned es_type = shader->key.part.gs.es->type;
+ gl_shader_stage es_stage = shader->key.part.gs.es->info.stage;
unsigned es_vgpr_comp_cnt, gs_vgpr_comp_cnt;
- if (es_type == PIPE_SHADER_VERTEX) {
+ if (es_stage == MESA_SHADER_VERTEX) {
es_vgpr_comp_cnt = si_get_vs_vgpr_comp_cnt(sscreen, shader, false);
- } else if (es_type == PIPE_SHADER_TESS_EVAL)
+ } else if (es_stage == MESA_SHADER_TESS_EVAL)
es_vgpr_comp_cnt = shader->key.part.gs.es->info.uses_primid ? 3 : 2;
else
unreachable("invalid shader selector type");
gs_vgpr_comp_cnt = 0; /* VGPR0 contains offsets 0, 1 */
unsigned num_user_sgprs;
- if (es_type == PIPE_SHADER_VERTEX)
+ if (es_stage == MESA_SHADER_VERTEX)
num_user_sgprs = si_get_num_vs_user_sgprs(shader, GFX9_VSGS_NUM_USER_SGPR);
else
num_user_sgprs = GFX9_TESGS_NUM_USER_SGPR;
S_00B228_GS_VGPR_COMP_CNT(gs_vgpr_comp_cnt);
uint32_t rsrc2 = S_00B22C_USER_SGPR(num_user_sgprs) |
S_00B22C_ES_VGPR_COMP_CNT(es_vgpr_comp_cnt) |
- S_00B22C_OC_LDS_EN(es_type == PIPE_SHADER_TESS_EVAL) |
+ S_00B22C_OC_LDS_EN(es_stage == MESA_SHADER_TESS_EVAL) |
S_00B22C_LDS_SIZE(shader->config.lds_size) |
S_00B22C_SCRATCH_EN(shader->config.scratch_bytes_per_wave > 0);
S_028A94_MAX_PRIMS_PER_SUBGROUP(shader->gs_info.max_prims_per_subgroup);
shader->ctx_reg.gs.vgt_esgs_ring_itemsize = shader->key.part.gs.es->esgs_itemsize / 4;
- if (es_type == PIPE_SHADER_TESS_EVAL)
+ if (es_stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader->key.part.gs.es, pm4);
polaris_set_vgt_vertex_reuse(sscreen, shader->key.part.gs.es, NULL, pm4);
unsigned si_get_input_prim(const struct si_shader_selector *gs)
{
- if (gs->type == PIPE_SHADER_GEOMETRY)
+ if (gs->info.stage == MESA_SHADER_GEOMETRY)
return gs->info.properties[TGSI_PROPERTY_GS_INPUT_PRIM];
- if (gs->type == PIPE_SHADER_TESS_EVAL) {
+ if (gs->info.stage == MESA_SHADER_TESS_EVAL) {
if (gs->info.properties[TGSI_PROPERTY_TES_POINT_MODE])
return PIPE_PRIM_POINTS;
if (gs->info.properties[TGSI_PROPERTY_TES_PRIM_MODE] == PIPE_PRIM_LINES)
{
const struct si_shader_selector *gs_sel = shader->selector;
const struct si_shader_info *gs_info = &gs_sel->info;
- enum pipe_shader_type gs_type = shader->selector->type;
+ const gl_shader_stage gs_stage = shader->selector->info.stage;
const struct si_shader_selector *es_sel =
shader->previous_stage_sel ? shader->previous_stage_sel : shader->selector;
const struct si_shader_info *es_info = &es_sel->info;
- enum pipe_shader_type es_type = es_sel->type;
+ const gl_shader_stage es_stage = es_sel->info.stage;
unsigned num_user_sgprs;
unsigned nparams, es_vgpr_comp_cnt, gs_vgpr_comp_cnt;
uint64_t va;
if (!pm4)
return;
- if (es_type == PIPE_SHADER_TESS_EVAL) {
- pm4->atom.emit = gs_type == PIPE_SHADER_GEOMETRY ? gfx10_emit_shader_ngg_tess_gs
+ if (es_stage == MESA_SHADER_TESS_EVAL) {
+ pm4->atom.emit = gs_stage == MESA_SHADER_GEOMETRY ? gfx10_emit_shader_ngg_tess_gs
: gfx10_emit_shader_ngg_tess_nogs;
} else {
- pm4->atom.emit = gs_type == PIPE_SHADER_GEOMETRY ? gfx10_emit_shader_ngg_notess_gs
+ pm4->atom.emit = gs_stage == MESA_SHADER_GEOMETRY ? gfx10_emit_shader_ngg_notess_gs
: gfx10_emit_shader_ngg_notess_nogs;
}
va = shader->bo->gpu_address;
- if (es_type == PIPE_SHADER_VERTEX) {
+ if (es_stage == MESA_SHADER_VERTEX) {
es_vgpr_comp_cnt = si_get_vs_vgpr_comp_cnt(sscreen, shader, false);
if (es_info->properties[TGSI_PROPERTY_VS_BLIT_SGPRS_AMD]) {
num_user_sgprs = si_get_num_vs_user_sgprs(shader, GFX9_VSGS_NUM_USER_SGPR);
}
} else {
- assert(es_type == PIPE_SHADER_TESS_EVAL);
+ assert(es_stage == MESA_SHADER_TESS_EVAL);
es_vgpr_comp_cnt = es_enable_prim_id ? 3 : 2;
num_user_sgprs = GFX9_TESGS_NUM_USER_SGPR;
* for the GL_LINE polygon mode to skip rendering lines on inner edges.
*/
if (gs_info->uses_invocationid ||
- (gs_type == PIPE_SHADER_VERTEX && !gfx10_is_ngg_passthrough(shader)))
+ (gs_stage == MESA_SHADER_VERTEX && !gfx10_is_ngg_passthrough(shader)))
gs_vgpr_comp_cnt = 3; /* VGPR3 contains InvocationID, edge flags. */
- else if ((gs_type == PIPE_SHADER_GEOMETRY && gs_info->uses_primid) ||
- (gs_type == PIPE_SHADER_VERTEX && shader->key.mono.u.vs_export_prim_id))
+ else if ((gs_stage == MESA_SHADER_GEOMETRY && gs_info->uses_primid) ||
+ (gs_stage == MESA_SHADER_VERTEX && shader->key.mono.u.vs_export_prim_id))
gs_vgpr_comp_cnt = 2; /* VGPR2 contains PrimitiveID. */
else if (input_prim >= PIPE_PRIM_TRIANGLES && !gfx10_is_ngg_passthrough(shader))
gs_vgpr_comp_cnt = 1; /* VGPR1 contains offsets 2, 3 */
S_00B22C_USER_SGPR(num_user_sgprs) |
S_00B22C_ES_VGPR_COMP_CNT(es_vgpr_comp_cnt) |
S_00B22C_USER_SGPR_MSB_GFX10(num_user_sgprs >> 5) |
- S_00B22C_OC_LDS_EN(es_type == PIPE_SHADER_TESS_EVAL) |
+ S_00B22C_OC_LDS_EN(es_stage == MESA_SHADER_TESS_EVAL) |
S_00B22C_LDS_SIZE(shader->config.lds_size));
/* Determine LATE_ALLOC_GS. */
S_028A84_NGG_DISABLE_PROVOK_REUSE(shader->key.mono.u.vs_export_prim_id ||
gs_sel->info.writes_primid);
- if (gs_type == PIPE_SHADER_GEOMETRY) {
+ if (gs_stage == MESA_SHADER_GEOMETRY) {
shader->ctx_reg.ngg.vgt_esgs_ring_itemsize = es_sel->esgs_itemsize / 4;
shader->ctx_reg.ngg.vgt_gs_max_vert_out = gs_sel->gs_max_out_vertices;
} else {
shader->ctx_reg.ngg.vgt_esgs_ring_itemsize = 1;
}
- if (es_type == PIPE_SHADER_TESS_EVAL)
+ if (es_stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, es_sel, pm4);
shader->ctx_reg.ngg.vgt_gs_onchip_cntl =
* this.
*/
shader->ctx_reg.ngg.pa_cl_ngg_cntl =
- S_028838_INDEX_BUF_EDGE_FLAG_ENA(gs_type == PIPE_SHADER_VERTEX) |
+ S_028838_INDEX_BUF_EDGE_FLAG_ENA(gs_stage == MESA_SHADER_VERTEX) |
/* Reuse for NGG. */
S_028838_VERTEX_REUSE_DEPTH(sscreen->info.chip_class >= GFX10_3 ? 30 : 0);
shader->pa_cl_vs_out_cntl = si_get_vs_out_cntl(gs_sel, true);
* Requirement: GE_CNTL.VERT_GRP_SIZE = VGT_GS_ONCHIP_CNTL.ES_VERTS_PER_SUBGRP - 5
*/
if ((sscreen->info.chip_class == GFX10) &&
- (es_type == PIPE_SHADER_VERTEX || gs_type == PIPE_SHADER_VERTEX) && /* = no tess */
+ (es_stage == MESA_SHADER_VERTEX || gs_stage == MESA_SHADER_VERTEX) && /* = no tess */
shader->ngg.hw_max_esverts != 256) {
shader->ge_cntl &= C_03096C_VERT_GRP_SIZE;
radeon_opt_set_context_reg(sctx, R_028818_PA_CL_VTE_CNTL, SI_TRACKED_PA_CL_VTE_CNTL,
shader->ctx_reg.vs.pa_cl_vte_cntl);
- if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
+ if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL)
radeon_opt_set_context_reg(sctx, R_028B6C_VGT_TF_PARAM, SI_TRACKED_VGT_TF_PARAM,
shader->vgt_tf_param);
shader->vgt_vertex_reuse_block_cntl);
/* Required programming for tessellation. (legacy pipeline only) */
- if (sctx->chip_class >= GFX10 && shader->selector->type == PIPE_SHADER_TESS_EVAL) {
+ if (sctx->chip_class >= GFX10 && shader->selector->info.stage == MESA_SHADER_TESS_EVAL) {
radeon_opt_set_context_reg(sctx, R_028A44_VGT_GS_ONCHIP_CNTL,
SI_TRACKED_VGT_GS_ONCHIP_CNTL,
S_028A44_ES_VERTS_PER_SUBGRP(250) |
if (gs) {
vgpr_comp_cnt = 0; /* only VertexID is needed for GS-COPY. */
num_user_sgprs = SI_GSCOPY_NUM_USER_SGPR;
- } else if (shader->selector->type == PIPE_SHADER_VERTEX) {
+ } else if (shader->selector->info.stage == MESA_SHADER_VERTEX) {
vgpr_comp_cnt = si_get_vs_vgpr_comp_cnt(sscreen, shader, enable_prim_id);
if (info->properties[TGSI_PROPERTY_VS_BLIT_SGPRS_AMD]) {
} else {
num_user_sgprs = si_get_num_vs_user_sgprs(shader, SI_VS_NUM_USER_SGPR);
}
- } else if (shader->selector->type == PIPE_SHADER_TESS_EVAL) {
+ } else if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL) {
vgpr_comp_cnt = enable_prim_id ? 3 : 2;
num_user_sgprs = SI_TES_NUM_USER_SGPR;
} else
S_030980_NUM_PC_LINES(sscreen->info.pc_lines / 4 - 1);
shader->pa_cl_vs_out_cntl = si_get_vs_out_cntl(shader->selector, false);
- oc_lds_en = shader->selector->type == PIPE_SHADER_TESS_EVAL ? 1 : 0;
+ oc_lds_en = shader->selector->info.stage == MESA_SHADER_TESS_EVAL ? 1 : 0;
si_pm4_set_reg(pm4, R_00B120_SPI_SHADER_PGM_LO_VS, va >> 8);
si_pm4_set_reg(pm4, R_00B124_SPI_SHADER_PGM_HI_VS, S_00B124_MEM_BASE(va >> 40));
S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1);
- if (shader->selector->type == PIPE_SHADER_TESS_EVAL)
+ if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader->selector, pm4);
polaris_set_vgt_vertex_reuse(sscreen, shader->selector, shader, pm4);
static void si_shader_init_pm4_state(struct si_screen *sscreen, struct si_shader *shader)
{
- switch (shader->selector->type) {
- case PIPE_SHADER_VERTEX:
+ switch (shader->selector->info.stage) {
+ case MESA_SHADER_VERTEX:
if (shader->key.as_ls)
si_shader_ls(sscreen, shader);
else if (shader->key.as_es)
else
si_shader_vs(sscreen, shader, NULL);
break;
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
si_shader_hs(sscreen, shader);
break;
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_TESS_EVAL:
if (shader->key.as_es)
si_shader_es(sscreen, shader);
else if (shader->key.as_ngg)
else
si_shader_vs(sscreen, shader, NULL);
break;
- case PIPE_SHADER_GEOMETRY:
+ case MESA_SHADER_GEOMETRY:
if (shader->key.as_ngg)
gfx10_shader_ngg(sscreen, shader);
else
si_shader_gs(sscreen, shader);
break;
- case PIPE_SHADER_FRAGMENT:
+ case MESA_SHADER_FRAGMENT:
si_shader_ps(sscreen, shader);
break;
default:
memset(key, 0, sizeof(*key));
- switch (sel->type) {
- case PIPE_SHADER_VERTEX:
+ switch (sel->info.stage) {
+ case MESA_SHADER_VERTEX:
si_shader_selector_key_vs(sctx, sel, key, &key->part.vs.prolog);
if (sctx->tes_shader.cso)
key->mono.u.vs_export_prim_id = 1;
}
break;
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
if (sctx->chip_class >= GFX9) {
si_shader_selector_key_vs(sctx, sctx->vs_shader.cso, key, &key->part.tcs.ls_prolog);
key->part.tcs.ls = sctx->vs_shader.cso;
if (sel == sctx->fixed_func_tcs_shader.cso)
key->mono.u.ff_tcs_inputs_to_copy = sctx->vs_shader.cso->outputs_written;
break;
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_TESS_EVAL:
key->as_ngg = stages_key.u.ngg;
if (sctx->gs_shader.cso)
key->mono.u.vs_export_prim_id = 1;
}
break;
- case PIPE_SHADER_GEOMETRY:
+ case MESA_SHADER_GEOMETRY:
if (sctx->chip_class >= GFX9) {
if (sctx->tes_shader.cso) {
key->part.gs.es = sctx->tes_shader.cso;
}
key->part.gs.prolog.tri_strip_adj_fix = sctx->gs_tri_strip_adj_fix;
break;
- case PIPE_SHADER_FRAGMENT: {
+ case MESA_SHADER_FRAGMENT: {
struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
struct si_state_blend *blend = sctx->queued.named.blend;
si_init_compiler(sscreen, compiler);
if (unlikely(!si_create_shader_variant(sscreen, compiler, shader, debug))) {
- PRINT_ERR("Failed to build shader variant (type=%u)\n", sel->type);
+ PRINT_ERR("Failed to build shader variant (type=%u)\n", sel->info.stage);
shader->compilation_failed = true;
return;
}
/* If this is a merged shader, get the first shader's selector. */
if (sscreen->info.chip_class >= GFX9) {
- if (sel->type == PIPE_SHADER_TESS_CTRL)
+ if (sel->info.stage == MESA_SHADER_TESS_CTRL)
previous_stage_sel = key->part.tcs.ls;
- else if (sel->type == PIPE_SHADER_GEOMETRY)
+ else if (sel->info.stage == MESA_SHADER_GEOMETRY)
previous_stage_sel = key->part.gs.es;
/* We need to wait for the previous shader. */
if (previous_stage_sel) {
struct si_shader_key shader1_key = zeroed;
- if (sel->type == PIPE_SHADER_TESS_CTRL) {
+ if (sel->info.stage == MESA_SHADER_TESS_CTRL) {
shader1_key.as_ls = 1;
- } else if (sel->type == PIPE_SHADER_GEOMETRY) {
+ } else if (sel->info.stage == MESA_SHADER_GEOMETRY) {
shader1_key.as_es = 1;
shader1_key.as_ngg = key->as_ngg; /* for Wave32 vs Wave64 */
} else {
{
unsigned next_shader = info->properties[TGSI_PROPERTY_NEXT_SHADER];
- switch (info->processor) {
- case PIPE_SHADER_VERTEX:
+ switch (info->stage) {
+ case MESA_SHADER_VERTEX:
switch (next_shader) {
case PIPE_SHADER_GEOMETRY:
key->as_es = 1;
}
break;
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_TESS_EVAL:
if (next_shader == PIPE_SHADER_GEOMETRY || !info->writes_position)
key->as_es = 1;
break;
+
+ default:;
}
}
si_parse_next_shader_property(&sel->info, sel->so.num_outputs != 0, &shader->key);
if (sscreen->use_ngg && (!sel->so.num_outputs || sscreen->use_ngg_streamout) &&
- ((sel->type == PIPE_SHADER_VERTEX && !shader->key.as_ls) ||
- sel->type == PIPE_SHADER_TESS_EVAL || sel->type == PIPE_SHADER_GEOMETRY))
+ ((sel->info.stage == MESA_SHADER_VERTEX && !shader->key.as_ls) ||
+ sel->info.stage == MESA_SHADER_TESS_EVAL || sel->info.stage == MESA_SHADER_GEOMETRY))
shader->key.as_ngg = 1;
if (sel->nir) {
*
* This is only done if non-monolithic shaders are enabled.
*/
- if ((sel->type == PIPE_SHADER_VERTEX || sel->type == PIPE_SHADER_TESS_EVAL) &&
+ if ((sel->info.stage == MESA_SHADER_VERTEX || sel->info.stage == MESA_SHADER_TESS_EVAL) &&
!shader->key.as_ls && !shader->key.as_es) {
unsigned i;
}
/* The GS copy shader is always pre-compiled. */
- if (sel->type == PIPE_SHADER_GEOMETRY &&
+ if (sel->info.stage == MESA_SHADER_GEOMETRY &&
(!sscreen->use_ngg || !sscreen->use_ngg_streamout || /* also for PRIMITIVES_GENERATED */
sel->tess_turns_off_ngg)) {
sel->gs_copy_shader = si_generate_gs_copy_shader(sscreen, compiler, sel, debug);
}
}
-void si_schedule_initial_compile(struct si_context *sctx, unsigned processor,
+void si_schedule_initial_compile(struct si_context *sctx, gl_shader_stage stage,
struct util_queue_fence *ready_fence,
struct si_compiler_ctx_state *compiler_ctx_state, void *job,
util_queue_execute_func execute)
struct util_async_debug_callback async_debug;
bool debug = (sctx->debug.debug_message && !sctx->debug.async) || sctx->is_debug ||
- si_can_dump_shader(sctx->screen, processor);
+ si_can_dump_shader(sctx->screen, pipe_shader_type_from_mesa(stage));
if (debug) {
u_async_debug_init(&async_debug);
si_nir_scan_shader(sel->nir, &sel->info);
- sel->type = sel->info.processor;
+ sel->type = pipe_shader_type_from_mesa(sel->info.stage);
p_atomic_inc(&sscreen->num_shaders_created);
si_get_active_slot_masks(&sel->info, &sel->active_const_and_shader_buffers,
&sel->active_samplers_and_images);
}
sel->num_vs_inputs =
- sel->type == PIPE_SHADER_VERTEX && !sel->info.properties[TGSI_PROPERTY_VS_BLIT_SGPRS_AMD]
+ sel->info.stage == MESA_SHADER_VERTEX && !sel->info.properties[TGSI_PROPERTY_VS_BLIT_SGPRS_AMD]
? sel->info.num_inputs
: 0;
sel->num_vbos_in_user_sgprs = MIN2(sel->num_vs_inputs, sscreen->num_vbos_in_user_sgprs);
/* The prolog is a no-op if there are no inputs. */
- sel->vs_needs_prolog = sel->type == PIPE_SHADER_VERTEX && sel->info.num_inputs &&
+ sel->vs_needs_prolog = sel->info.stage == MESA_SHADER_VERTEX && sel->info.num_inputs &&
!sel->info.properties[TGSI_PROPERTY_VS_BLIT_SGPRS_AMD];
sel->prim_discard_cs_allowed =
- sel->type == PIPE_SHADER_VERTEX && !sel->info.uses_bindless_images &&
+ sel->info.stage == MESA_SHADER_VERTEX && !sel->info.uses_bindless_images &&
!sel->info.uses_bindless_samplers && !sel->info.writes_memory &&
!sel->info.writes_viewport_index &&
!sel->info.properties[TGSI_PROPERTY_VS_WINDOW_SPACE_POSITION] && !sel->so.num_outputs;
- switch (sel->type) {
- case PIPE_SHADER_GEOMETRY:
+ switch (sel->info.stage) {
+ case MESA_SHADER_GEOMETRY:
sel->gs_output_prim = sel->info.properties[TGSI_PROPERTY_GS_OUTPUT_PRIM];
/* Only possibilities: POINTS, LINE_STRIP, TRIANGLES */
(sel->info.num_outputs * 4 + 1) > 6500 /* max dw per GS primitive */);
break;
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
/* Always reserve space for these. */
sel->patch_outputs_written |=
(1ull << si_shader_io_get_unique_index_patch(TGSI_SEMANTIC_TESSINNER, 0)) |
(1ull << si_shader_io_get_unique_index_patch(TGSI_SEMANTIC_TESSOUTER, 0));
/* fall through */
- case PIPE_SHADER_VERTEX:
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_VERTEX:
+ case MESA_SHADER_TESS_EVAL:
for (i = 0; i < sel->info.num_outputs; i++) {
unsigned name = sel->info.output_semantic_name[i];
unsigned index = sel->info.output_semantic_index[i];
sel->rast_prim = PIPE_PRIM_TRIANGLES;
break;
- case PIPE_SHADER_FRAGMENT:
+ case MESA_SHADER_FRAGMENT:
for (i = 0; i < sel->info.num_inputs; i++) {
unsigned name = sel->info.input_semantic_name[i];
unsigned index = sel->info.input_semantic_index[i];
sscreen->info.chip_class >= GFX10 &&
sscreen->info.has_dedicated_vram &&
sscreen->use_ngg_culling &&
- (sel->type == PIPE_SHADER_VERTEX ||
- (sel->type == PIPE_SHADER_TESS_EVAL &&
+ (sel->info.stage == MESA_SHADER_VERTEX ||
+ (sel->info.stage == MESA_SHADER_TESS_EVAL &&
(sscreen->always_use_ngg_culling_all ||
sscreen->always_use_ngg_culling_tess))) &&
sel->info.writes_position &&
(void)simple_mtx_init(&sel->mutex, mtx_plain);
- si_schedule_initial_compile(sctx, sel->info.processor, &sel->ready, &sel->compiler_ctx_state,
+ si_schedule_initial_compile(sctx, sel->info.stage, &sel->ready, &sel->compiler_ctx_state,
sel, si_init_shader_selector_async);
return sel;
}
* binding it to the same shader stage would be considered
* a no-op, causing random behavior.
*/
- switch (shader->selector->type) {
- case PIPE_SHADER_VERTEX:
+ switch (shader->selector->info.stage) {
+ case MESA_SHADER_VERTEX:
if (shader->key.as_ls) {
assert(sctx->chip_class <= GFX8);
si_pm4_delete_state(sctx, ls, shader->pm4);
si_pm4_delete_state(sctx, vs, shader->pm4);
}
break;
- case PIPE_SHADER_TESS_CTRL:
+ case MESA_SHADER_TESS_CTRL:
si_pm4_delete_state(sctx, hs, shader->pm4);
break;
- case PIPE_SHADER_TESS_EVAL:
+ case MESA_SHADER_TESS_EVAL:
if (shader->key.as_es) {
assert(sctx->chip_class <= GFX8);
si_pm4_delete_state(sctx, es, shader->pm4);
si_pm4_delete_state(sctx, vs, shader->pm4);
}
break;
- case PIPE_SHADER_GEOMETRY:
+ case MESA_SHADER_GEOMETRY:
if (shader->is_gs_copy_shader)
si_pm4_delete_state(sctx, vs, shader->pm4);
else
si_pm4_delete_state(sctx, gs, shader->pm4);
break;
- case PIPE_SHADER_FRAGMENT:
+ case MESA_SHADER_FRAGMENT:
si_pm4_delete_state(sctx, ps, shader->pm4);
break;
default:;
struct si_shader_selector *sel = (struct si_shader_selector *)cso;
struct si_shader *p = sel->first_variant, *c;
struct si_shader_ctx_state *current_shader[SI_NUM_SHADERS] = {
- [PIPE_SHADER_VERTEX] = &sctx->vs_shader, [PIPE_SHADER_TESS_CTRL] = &sctx->tcs_shader,
- [PIPE_SHADER_TESS_EVAL] = &sctx->tes_shader, [PIPE_SHADER_GEOMETRY] = &sctx->gs_shader,
- [PIPE_SHADER_FRAGMENT] = &sctx->ps_shader,
+ [MESA_SHADER_VERTEX] = &sctx->vs_shader,
+ [MESA_SHADER_TESS_CTRL] = &sctx->tcs_shader,
+ [MESA_SHADER_TESS_EVAL] = &sctx->tes_shader,
+ [MESA_SHADER_GEOMETRY] = &sctx->gs_shader,
+ [MESA_SHADER_FRAGMENT] = &sctx->ps_shader,
};
util_queue_drop_job(&sctx->screen->shader_compiler_queue, &sel->ready);
- if (current_shader[sel->type]->cso == sel) {
- current_shader[sel->type]->cso = NULL;
- current_shader[sel->type]->current = NULL;
+ if (current_shader[sel->info.stage]->cso == sel) {
+ current_shader[sel->info.stage]->cso = NULL;
+ current_shader[sel->info.stage]->current = NULL;
}
while (p) {
projects / mesa.git / commitdiff