fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-21 22:20:14 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

radeonsi: put si_pm4_state at the beginning of si_shader

Browse source 
instead of allocating it separately. This removes pointer indirections.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12343>
This commit is contained in:
Marek Olšák 2021-08-12 09:39:04 -04:00 • committed by Marge Bot
parent 73c82570cb
commit 3df035d08c
6 changed files with 107 additions and 119 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
src/gallium/drivers/radeonsi/si_pm4.c
View file

@ -117,8 +117,8 @@ void si_pm4_emit(struct si_context *sctx, struct si_pm4_state *state)
{ {
struct radeon_cmdbuf *cs = &sctx->gfx_cs; struct radeon_cmdbuf *cs = &sctx->gfx_cs;
if (state->shader) { if (state->is_shader) {
radeon_add_to_buffer_list(sctx, &sctx->gfx_cs, state->shader->bo, radeon_add_to_buffer_list(sctx, &sctx->gfx_cs, ((struct si_shader*)state)->bo,
RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY); RADEON_USAGE_READ, RADEON_PRIO_SHADER_BINARY);
} }
@ -139,7 +139,7 @@ void si_pm4_reset_emitted(struct si_context *sctx, bool first_cs)
for (unsigned i = 0; i < SI_NUM_STATES; i++) { for (unsigned i = 0; i < SI_NUM_STATES; i++) {
struct si_pm4_state *state = sctx->emitted.array[i]; struct si_pm4_state *state = sctx->emitted.array[i];
if (state && state->shader) { if (state && state->is_shader) {
sctx->emitted.array[i] = NULL; sctx->emitted.array[i] = NULL;
sctx->dirty_states |= 1 << i; sctx->dirty_states |= 1 << i;
} }

2
src/gallium/drivers/radeonsi/si_pm4.h
View file

@ -54,7 +54,7 @@ struct si_pm4_state {
uint32_t pm4[SI_PM4_MAX_DW]; uint32_t pm4[SI_PM4_MAX_DW];
/* For shader states only */ /* For shader states only */
struct si_shader *shader; bool is_shader;
struct si_atom atom; struct si_atom atom;
}; };

3
src/gallium/drivers/radeonsi/si_shader.h
View file

@ -138,6 +138,7 @@
#include "util/u_inlines.h" #include "util/u_inlines.h"
#include "util/u_live_shader_cache.h" #include "util/u_live_shader_cache.h"
#include "util/u_queue.h" #include "util/u_queue.h"
#include "si_pm4.h"
#include <stdio.h> #include <stdio.h>
@ -772,6 +773,7 @@ union si_vgt_stages_key {
}; };
struct si_shader { struct si_shader {
struct si_pm4_state pm4; /* base class */
struct si_compiler_ctx_state compiler_ctx_state; struct si_compiler_ctx_state compiler_ctx_state;
struct si_shader_selector *selector; struct si_shader_selector *selector;
@ -783,7 +785,6 @@ struct si_shader {
struct si_shader_part *prolog2; struct si_shader_part *prolog2;
struct si_shader_part *epilog; struct si_shader_part *epilog;
struct si_pm4_state *pm4;
struct si_resource *bo; struct si_resource *bo;
struct si_resource *scratch_bo; struct si_resource *scratch_bo;
struct si_shader_key key; struct si_shader_key key;

12
src/gallium/drivers/radeonsi/si_state.h
View file

@ -183,13 +183,13 @@ union si_state {
struct si_state_rasterizer *rasterizer; struct si_state_rasterizer *rasterizer;
struct si_state_dsa *dsa; struct si_state_dsa *dsa;
struct si_pm4_state *poly_offset; struct si_pm4_state *poly_offset;
struct si_pm4_state *ls; struct si_shader *ls;
struct si_pm4_state *hs; struct si_shader *hs;
struct si_pm4_state *es; struct si_shader *es;
struct si_pm4_state *gs; struct si_shader *gs;
struct si_pm4_state *vgt_shader_config; struct si_pm4_state *vgt_shader_config;
struct si_pm4_state *vs; struct si_shader *vs;
struct si_pm4_state *ps; struct si_shader *ps;
} named; } named;
struct si_pm4_state *array[sizeof(struct si_state_named) / sizeof(struct si_pm4_state *)]; struct si_pm4_state *array[sizeof(struct si_state_named) / sizeof(struct si_pm4_state *)];
}; };

32
src/gallium/drivers/radeonsi/si_state_draw.cpp
View file

@ -127,7 +127,7 @@ static bool si_update_shaders(struct si_context *sctx)
r = si_shader_select(ctx, &sctx->shader.tcs); r = si_shader_select(ctx, &sctx->shader.tcs);
if (r) if (r)
return false; return false;
si_pm4_bind_state(sctx, hs, sctx->shader.tcs.current->pm4); si_pm4_bind_state(sctx, hs, sctx->shader.tcs.current);
} else { } else {
if (!sctx->fixed_func_tcs_shader.cso) { if (!sctx->fixed_func_tcs_shader.cso) {
sctx->fixed_func_tcs_shader.cso = sctx->fixed_func_tcs_shader.cso =
@ -142,7 +142,7 @@ static bool si_update_shaders(struct si_context *sctx)
r = si_shader_select(ctx, &sctx->fixed_func_tcs_shader); r = si_shader_select(ctx, &sctx->fixed_func_tcs_shader);
if (r) if (r)
return false; return false;
si_pm4_bind_state(sctx, hs, sctx->fixed_func_tcs_shader.current->pm4); si_pm4_bind_state(sctx, hs, sctx->fixed_func_tcs_shader.current);
} }
if (!HAS_GS || GFX_VERSION <= GFX8) { if (!HAS_GS || GFX_VERSION <= GFX8) {
@ -153,11 +153,11 @@ static bool si_update_shaders(struct si_context *sctx)
if (HAS_GS) { if (HAS_GS) {
/* TES as ES */ /* TES as ES */
assert(GFX_VERSION <= GFX8); assert(GFX_VERSION <= GFX8);
si_pm4_bind_state(sctx, es, sctx->shader.tes.current->pm4); si_pm4_bind_state(sctx, es, sctx->shader.tes.current);
} else if (NGG) { } else if (NGG) {
si_pm4_bind_state(sctx, gs, sctx->shader.tes.current->pm4); si_pm4_bind_state(sctx, gs, sctx->shader.tes.current);
} else { } else {
si_pm4_bind_state(sctx, vs, sctx->shader.tes.current->pm4); si_pm4_bind_state(sctx, vs, sctx->shader.tes.current);
} }
} }
} else { } else {
@ -174,9 +174,9 @@ static bool si_update_shaders(struct si_context *sctx)
r = si_shader_select(ctx, &sctx->shader.gs); r = si_shader_select(ctx, &sctx->shader.gs);
if (r) if (r)
return false; return false;
si_pm4_bind_state(sctx, gs, sctx->shader.gs.current->pm4); si_pm4_bind_state(sctx, gs, sctx->shader.gs.current);
if (!NGG) { if (!NGG) {
si_pm4_bind_state(sctx, vs, sctx->shader.gs.cso->gs_copy_shader->pm4); si_pm4_bind_state(sctx, vs, sctx->shader.gs.cso->gs_copy_shader);
if (!si_update_gs_ring_buffers(sctx)) if (!si_update_gs_ring_buffers(sctx))
return false; return false;
@ -203,23 +203,23 @@ static bool si_update_shaders(struct si_context *sctx)
if (!HAS_TESS && !HAS_GS) { if (!HAS_TESS && !HAS_GS) {
if (NGG) { if (NGG) {
si_pm4_bind_state(sctx, gs, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, gs, sctx->shader.vs.current);
si_pm4_bind_state(sctx, vs, NULL); si_pm4_bind_state(sctx, vs, NULL);
sctx->prefetch_L2_mask &= ~SI_PREFETCH_VS; sctx->prefetch_L2_mask &= ~SI_PREFETCH_VS;
} else { } else {
si_pm4_bind_state(sctx, vs, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, vs, sctx->shader.vs.current);
} }
} else if (HAS_TESS) { } else if (HAS_TESS) {
si_pm4_bind_state(sctx, ls, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, ls, sctx->shader.vs.current);
} else { } else {
assert(HAS_GS); assert(HAS_GS);
si_pm4_bind_state(sctx, es, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, es, sctx->shader.vs.current);
} }
} }
sctx->vs_uses_base_instance = sctx->vs_uses_base_instance =
sctx->shader.vs.current ? sctx->shader.vs.current->uses_base_instance : sctx->shader.vs.current ? sctx->shader.vs.current->uses_base_instance :
sctx->queued.named.hs ? sctx->queued.named.hs->shader->uses_base_instance : sctx->queued.named.hs ? sctx->queued.named.hs->uses_base_instance :
sctx->shader.gs.current->uses_base_instance; sctx->shader.gs.current->uses_base_instance;
union si_vgt_stages_key key; union si_vgt_stages_key key;
@ -244,7 +244,7 @@ static bool si_update_shaders(struct si_context *sctx)
r = si_shader_select(ctx, &sctx->shader.ps); r = si_shader_select(ctx, &sctx->shader.ps);
if (r) if (r)
return false; return false;
si_pm4_bind_state(sctx, ps, sctx->shader.ps.current->pm4); si_pm4_bind_state(sctx, ps, sctx->shader.ps.current);
if (si_pm4_state_changed(sctx, ps) || if (si_pm4_state_changed(sctx, ps) ||
(!NGG && si_pm4_state_changed(sctx, vs)) || (!NGG && si_pm4_state_changed(sctx, vs)) ||
@ -310,7 +310,7 @@ static bool si_update_shaders(struct si_context *sctx)
if (GFX_VERSION <= GFX8) /* LS */ if (GFX_VERSION <= GFX8) /* LS */
scratch_size = MAX2(scratch_size, sctx->shader.vs.current->config.scratch_bytes_per_wave); scratch_size = MAX2(scratch_size, sctx->shader.vs.current->config.scratch_bytes_per_wave);
scratch_size = MAX2(scratch_size, sctx->queued.named.hs->shader->config.scratch_bytes_per_wave); scratch_size = MAX2(scratch_size, sctx->queued.named.hs->config.scratch_bytes_per_wave);
if (HAS_GS) { if (HAS_GS) {
if (GFX_VERSION <= GFX8) /* ES */ if (GFX_VERSION <= GFX8) /* ES */
@ -383,9 +383,9 @@ static unsigned si_conv_pipe_prim(unsigned mode)
return prim_conv[mode]; return prim_conv[mode];
} }
static void si_prefetch_shader_async(struct si_context *sctx, struct si_pm4_state *state) static void si_prefetch_shader_async(struct si_context *sctx, struct si_shader *shader)
{ {
struct pipe_resource *bo = &state->shader->bo->b.b; struct pipe_resource *bo = &shader->bo->b.b;
si_cp_dma_prefetch(sctx, bo, 0, bo->width0); si_cp_dma_prefetch(sctx, bo, 0, bo->width0);
} }

171
src/gallium/drivers/radeonsi/si_state_shaders.c
View file

@ -373,7 +373,7 @@ bool si_shader_mem_ordered(struct si_shader *shader)
} }
static void si_set_tesseval_regs(struct si_screen *sscreen, const struct si_shader_selector *tes, static void si_set_tesseval_regs(struct si_screen *sscreen, const struct si_shader_selector *tes,
struct si_pm4_state *pm4) struct si_shader *shader)
{ {
const struct si_shader_info *info = &tes->info; const struct si_shader_info *info = &tes->info;
unsigned tes_prim_mode = info->base.tess.primitive_mode; unsigned tes_prim_mode = info->base.tess.primitive_mode;
@ -430,10 +430,9 @@ static void si_set_tesseval_regs(struct si_screen *sscreen, const struct si_shad
} else } else
distribution_mode = V_028B6C_NO_DIST; distribution_mode = V_028B6C_NO_DIST;
assert(pm4->shader); shader->vgt_tf_param = S_028B6C_TYPE(type) | S_028B6C_PARTITIONING(partitioning) |
pm4->shader->vgt_tf_param = S_028B6C_TYPE(type) | S_028B6C_PARTITIONING(partitioning) | S_028B6C_TOPOLOGY(topology) |
S_028B6C_TOPOLOGY(topology) | S_028B6C_DISTRIBUTION_MODE(distribution_mode);
S_028B6C_DISTRIBUTION_MODE(distribution_mode);
} }
/* Polaris needs different VTX_REUSE_DEPTH settings depending on /* Polaris needs different VTX_REUSE_DEPTH settings depending on
@ -447,18 +446,16 @@ static void si_set_tesseval_regs(struct si_screen *sscreen, const struct si_shad
* VS as ES | ES -> GS -> VS | 30 * VS as ES | ES -> GS -> VS | 30
* TES as VS | LS -> HS -> VS | 14 or 30 * TES as VS | LS -> HS -> VS | 14 or 30
* TES as ES | LS -> HS -> ES -> GS -> VS | 14 or 30 * TES as ES | LS -> HS -> ES -> GS -> VS | 14 or 30
*
* If "shader" is NULL, it's assumed it's not LS or GS copy shader.
*/ */
static void polaris_set_vgt_vertex_reuse(struct si_screen *sscreen, struct si_shader_selector *sel, 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) struct si_shader *shader)
{ {
if (sscreen->info.family < CHIP_POLARIS10 || sscreen->info.chip_class >= GFX10) if (sscreen->info.family < CHIP_POLARIS10 || sscreen->info.chip_class >= GFX10)
return; return;
/* VS as VS, or VS as ES: */ /* VS as VS, or VS as ES: */
if ((sel->info.stage == MESA_SHADER_VERTEX && if ((sel->info.stage == MESA_SHADER_VERTEX &&
(!shader || (!shader->key.as_ls && !shader->is_gs_copy_shader))) || (!shader->key.as_ls && !shader->is_gs_copy_shader)) ||
/* TES as VS, or TES as ES: */ /* TES as VS, or TES as ES: */
sel->info.stage == MESA_SHADER_TESS_EVAL) { sel->info.stage == MESA_SHADER_TESS_EVAL) {
unsigned vtx_reuse_depth = 30; unsigned vtx_reuse_depth = 30;
@ -467,25 +464,15 @@ static void polaris_set_vgt_vertex_reuse(struct si_screen *sscreen, struct si_sh
sel->info.base.tess.spacing == TESS_SPACING_FRACTIONAL_ODD) sel->info.base.tess.spacing == TESS_SPACING_FRACTIONAL_ODD)
vtx_reuse_depth = 14; vtx_reuse_depth = 14;
assert(pm4->shader); shader->vgt_vertex_reuse_block_cntl = vtx_reuse_depth;
pm4->shader->vgt_vertex_reuse_block_cntl = vtx_reuse_depth;
} }
} }
static struct si_pm4_state *si_get_shader_pm4_state(struct si_shader *shader) static struct si_pm4_state *si_get_shader_pm4_state(struct si_shader *shader)
{ {
if (shader->pm4) si_pm4_clear_state(&shader->pm4);
si_pm4_clear_state(shader->pm4); shader->pm4.is_shader = true;
else return &shader->pm4;
shader->pm4 = CALLOC_STRUCT(si_pm4_state);
if (shader->pm4) {
shader->pm4->shader = shader;
return shader->pm4;
} else {
fprintf(stderr, "radeonsi: Failed to create pm4 state.\n");
return NULL;
}
} }
static unsigned si_get_num_vs_user_sgprs(struct si_shader *shader, static unsigned si_get_num_vs_user_sgprs(struct si_shader *shader,
@ -616,7 +603,7 @@ static void si_shader_hs(struct si_screen *sscreen, struct si_shader *shader)
static void si_emit_shader_es(struct si_context *sctx) static void si_emit_shader_es(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.es->shader; struct si_shader *shader = sctx->queued.named.es;
if (!shader) if (!shader)
return; return;
@ -677,9 +664,9 @@ static void si_shader_es(struct si_screen *sscreen, struct si_shader *shader)
S_00B32C_SCRATCH_EN(shader->config.scratch_bytes_per_wave > 0)); S_00B32C_SCRATCH_EN(shader->config.scratch_bytes_per_wave > 0));
if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL) if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader->selector, pm4); si_set_tesseval_regs(sscreen, shader->selector, shader);
polaris_set_vgt_vertex_reuse(sscreen, shader->selector, shader, pm4); polaris_set_vgt_vertex_reuse(sscreen, shader->selector, shader);
} }
void gfx9_get_gs_info(struct si_shader_selector *es, struct si_shader_selector *gs, void gfx9_get_gs_info(struct si_shader_selector *es, struct si_shader_selector *gs,
@ -777,7 +764,7 @@ void gfx9_get_gs_info(struct si_shader_selector *es, struct si_shader_selector *
static void si_emit_shader_gs(struct si_context *sctx) static void si_emit_shader_gs(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.gs->shader; struct si_shader *shader = sctx->queued.named.gs;
if (!shader) if (!shader)
return; return;
@ -953,9 +940,9 @@ static void si_shader_gs(struct si_screen *sscreen, struct si_shader *shader)
shader->ctx_reg.gs.vgt_esgs_ring_itemsize = shader->key.part.gs.es->esgs_itemsize / 4; shader->ctx_reg.gs.vgt_esgs_ring_itemsize = shader->key.part.gs.es->esgs_itemsize / 4;
if (es_stage == MESA_SHADER_TESS_EVAL) if (es_stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader->key.part.gs.es, pm4); si_set_tesseval_regs(sscreen, shader->key.part.gs.es, shader);
polaris_set_vgt_vertex_reuse(sscreen, shader->key.part.gs.es, NULL, pm4); polaris_set_vgt_vertex_reuse(sscreen, shader->key.part.gs.es, shader);
} else { } else {
if (sscreen->info.chip_class >= GFX7) { if (sscreen->info.chip_class >= GFX7) {
si_pm4_set_reg(pm4, R_00B21C_SPI_SHADER_PGM_RSRC3_GS, si_pm4_set_reg(pm4, R_00B21C_SPI_SHADER_PGM_RSRC3_GS,
@ -1046,7 +1033,7 @@ static void gfx10_emit_shader_ngg_tail(struct si_context *sctx, struct si_shader
static void gfx10_emit_shader_ngg_notess_nogs(struct si_context *sctx) static void gfx10_emit_shader_ngg_notess_nogs(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.gs->shader; struct si_shader *shader = sctx->queued.named.gs;
if (!shader) if (!shader)
return; return;
@ -1055,7 +1042,7 @@ static void gfx10_emit_shader_ngg_notess_nogs(struct si_context *sctx)
static void gfx10_emit_shader_ngg_tess_nogs(struct si_context *sctx) static void gfx10_emit_shader_ngg_tess_nogs(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.gs->shader; struct si_shader *shader = sctx->queued.named.gs;
if (!shader) if (!shader)
return; return;
@ -1069,7 +1056,7 @@ static void gfx10_emit_shader_ngg_tess_nogs(struct si_context *sctx)
static void gfx10_emit_shader_ngg_notess_gs(struct si_context *sctx) static void gfx10_emit_shader_ngg_notess_gs(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.gs->shader; struct si_shader *shader = sctx->queued.named.gs;
if (!shader) if (!shader)
return; return;
@ -1083,7 +1070,7 @@ static void gfx10_emit_shader_ngg_notess_gs(struct si_context *sctx)
static void gfx10_emit_shader_ngg_tess_gs(struct si_context *sctx) static void gfx10_emit_shader_ngg_tess_gs(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.gs->shader; struct si_shader *shader = sctx->queued.named.gs;
if (!shader) if (!shader)
return; return;
@ -1267,7 +1254,7 @@ static void gfx10_shader_ngg(struct si_screen *sscreen, struct si_shader *shader
} }
if (es_stage == MESA_SHADER_TESS_EVAL) if (es_stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, es_sel, pm4); si_set_tesseval_regs(sscreen, es_sel, shader);
shader->ctx_reg.ngg.vgt_gs_onchip_cntl = shader->ctx_reg.ngg.vgt_gs_onchip_cntl =
S_028A44_ES_VERTS_PER_SUBGRP(shader->ngg.hw_max_esverts) | S_028A44_ES_VERTS_PER_SUBGRP(shader->ngg.hw_max_esverts) |
@ -1356,7 +1343,7 @@ static void gfx10_shader_ngg(struct si_screen *sscreen, struct si_shader *shader
static void si_emit_shader_vs(struct si_context *sctx) static void si_emit_shader_vs(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.vs->shader; struct si_shader *shader = sctx->queued.named.vs;
if (!shader) if (!shader)
return; return;
@ -1556,9 +1543,9 @@ static void si_shader_vs(struct si_screen *sscreen, struct si_shader *shader,
S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1); S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1);
if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL) if (shader->selector->info.stage == MESA_SHADER_TESS_EVAL)
si_set_tesseval_regs(sscreen, shader->selector, pm4); si_set_tesseval_regs(sscreen, shader->selector, shader);
polaris_set_vgt_vertex_reuse(sscreen, shader->selector, shader, pm4); polaris_set_vgt_vertex_reuse(sscreen, shader->selector, shader);
} }
static unsigned si_get_ps_num_interp(struct si_shader *ps) static unsigned si_get_ps_num_interp(struct si_shader *ps)
@ -1593,7 +1580,7 @@ static unsigned si_get_spi_shader_col_format(struct si_shader *shader)
static void si_emit_shader_ps(struct si_context *sctx) static void si_emit_shader_ps(struct si_context *sctx)
{ {
struct si_shader *shader = sctx->queued.named.ps->shader; struct si_shader *shader = sctx->queued.named.ps;
if (!shader) if (!shader)
return; return;
@ -3414,55 +3401,55 @@ static void si_delete_shader(struct si_context *sctx, struct si_shader *shader)
util_queue_fence_destroy(&shader->ready); util_queue_fence_destroy(&shader->ready);
if (shader->pm4) { /* If destroyed shaders were not unbound, the next compiled
/* If destroyed shaders were not unbound, the next compiled * shader variant could get the same pointer address and so
* shader variant could get the same pointer address and so * binding it to the same shader stage would be considered
* binding it to the same shader stage would be considered * a no-op, causing random behavior.
* a no-op, causing random behavior. */
*/ int state_index = -1;
switch (shader->selector->info.stage) {
case MESA_SHADER_VERTEX: switch (shader->selector->info.stage) {
if (shader->key.as_ls) { case MESA_SHADER_VERTEX:
assert(sctx->chip_class <= GFX8); if (shader->key.as_ls) {
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(ls)); if (sctx->chip_class <= GFX8)
} else if (shader->key.as_es) { state_index = SI_STATE_IDX(ls);
assert(sctx->chip_class <= GFX8); } else if (shader->key.as_es) {
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(es)); if (sctx->chip_class <= GFX8)
} else if (shader->key.as_ngg) { state_index = SI_STATE_IDX(es);
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(gs)); } else if (shader->key.as_ngg) {
} else { state_index = SI_STATE_IDX(gs);
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(vs)); } else {
} state_index = SI_STATE_IDX(vs);
break;
case MESA_SHADER_TESS_CTRL:
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(hs));
break;
case MESA_SHADER_TESS_EVAL:
if (shader->key.as_es) {
assert(sctx->chip_class <= GFX8);
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(es));
} else if (shader->key.as_ngg) {
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(gs));
} else {
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(vs));
}
break;
case MESA_SHADER_GEOMETRY:
if (shader->is_gs_copy_shader)
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(vs));
else
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(gs));
break;
case MESA_SHADER_FRAGMENT:
si_pm4_free_state(sctx, shader->pm4, SI_STATE_IDX(ps));
break;
default:;
} }
break;
case MESA_SHADER_TESS_CTRL:
state_index = SI_STATE_IDX(hs);
break;
case MESA_SHADER_TESS_EVAL:
if (shader->key.as_es) {
if (sctx->chip_class <= GFX8)
state_index = SI_STATE_IDX(es);
} else if (shader->key.as_ngg) {
state_index = SI_STATE_IDX(gs);
} else {
state_index = SI_STATE_IDX(vs);
}
break;
case MESA_SHADER_GEOMETRY:
if (shader->is_gs_copy_shader)
state_index = SI_STATE_IDX(vs);
else
state_index = SI_STATE_IDX(gs);
break;
case MESA_SHADER_FRAGMENT:
state_index = SI_STATE_IDX(ps);
break;
default:;
} }
si_shader_selector_reference(sctx, &shader->previous_stage_sel, NULL); si_shader_selector_reference(sctx, &shader->previous_stage_sel, NULL);
si_shader_destroy(shader); si_shader_destroy(shader);
free(shader); si_pm4_free_state(sctx, &shader->pm4, state_index);
} }
static void si_destroy_shader_selector(struct pipe_context *ctx, void *cso) static void si_destroy_shader_selector(struct pipe_context *ctx, void *cso)
@ -3775,19 +3762,19 @@ static bool si_update_scratch_relocs(struct si_context *sctx)
if (r < 0) if (r < 0)
return false; return false;
if (r == 1) if (r == 1)
si_pm4_bind_state(sctx, ps, sctx->shader.ps.current->pm4); si_pm4_bind_state(sctx, ps, sctx->shader.ps.current);
r = si_update_scratch_buffer(sctx, sctx->shader.gs.current); r = si_update_scratch_buffer(sctx, sctx->shader.gs.current);
if (r < 0) if (r < 0)
return false; return false;
if (r == 1) if (r == 1)
si_pm4_bind_state(sctx, gs, sctx->shader.gs.current->pm4); si_pm4_bind_state(sctx, gs, sctx->shader.gs.current);
r = si_update_scratch_buffer(sctx, tcs); r = si_update_scratch_buffer(sctx, tcs);
if (r < 0) if (r < 0)
return false; return false;
if (r == 1) if (r == 1)
si_pm4_bind_state(sctx, hs, tcs->pm4); si_pm4_bind_state(sctx, hs, tcs);
/* VS can be bound as LS, ES, or VS. */ /* VS can be bound as LS, ES, or VS. */
r = si_update_scratch_buffer(sctx, sctx->shader.vs.current); r = si_update_scratch_buffer(sctx, sctx->shader.vs.current);
@ -3795,13 +3782,13 @@ static bool si_update_scratch_relocs(struct si_context *sctx)
return false; return false;
if (r == 1) { if (r == 1) {
if (sctx->shader.vs.current->key.as_ls) if (sctx->shader.vs.current->key.as_ls)
si_pm4_bind_state(sctx, ls, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, ls, sctx->shader.vs.current);
else if (sctx->shader.vs.current->key.as_es) else if (sctx->shader.vs.current->key.as_es)
si_pm4_bind_state(sctx, es, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, es, sctx->shader.vs.current);
else if (sctx->shader.vs.current->key.as_ngg) else if (sctx->shader.vs.current->key.as_ngg)
si_pm4_bind_state(sctx, gs, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, gs, sctx->shader.vs.current);
else else
si_pm4_bind_state(sctx, vs, sctx->shader.vs.current->pm4); si_pm4_bind_state(sctx, vs, sctx->shader.vs.current);
} }
/* TES can be bound as ES or VS. */ /* TES can be bound as ES or VS. */
@ -3810,11 +3797,11 @@ static bool si_update_scratch_relocs(struct si_context *sctx)
return false; return false;
if (r == 1) { if (r == 1) {
if (sctx->shader.tes.current->key.as_es) if (sctx->shader.tes.current->key.as_es)
si_pm4_bind_state(sctx, es, sctx->shader.tes.current->pm4); si_pm4_bind_state(sctx, es, sctx->shader.tes.current);
else if (sctx->shader.tes.current->key.as_ngg) else if (sctx->shader.tes.current->key.as_ngg)
si_pm4_bind_state(sctx, gs, sctx->shader.tes.current->pm4); si_pm4_bind_state(sctx, gs, sctx->shader.tes.current);
else else
si_pm4_bind_state(sctx, vs, sctx->shader.tes.current->pm4); si_pm4_bind_state(sctx, vs, sctx->shader.tes.current);
} }
return true; return true;