KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

radeonsi: don't use an indirect table for state atoms 

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
This commit is contained in:
Marek Olšák 2018-04-08 20:54:02 -04:00
parent 9054799b39
commit 639b673fc3
14 changed files with 147 additions and 171 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/gallium/drivers/radeonsi/si_blit.c
View File

@ -85,7 +85,7 @@ void si_blitter_end(struct si_context *sctx)
* non-global VS user SGPRs. */ * non-global VS user SGPRs. */
sctx->shader_pointers_dirty |= SI_DESCS_SHADER_MASK(VERTEX); sctx->shader_pointers_dirty |= SI_DESCS_SHADER_MASK(VERTEX);
sctx->vertex_buffer_pointer_dirty = sctx->vb_descriptors_buffer != NULL; sctx->vertex_buffer_pointer_dirty = sctx->vb_descriptors_buffer != NULL;
si_mark_atom_dirty(sctx, &sctx->shader_pointers.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.shader_pointers);
} }
static unsigned u_max_sample(struct pipe_resource *r) static unsigned u_max_sample(struct pipe_resource *r)
@ -109,7 +109,7 @@ si_blit_dbcb_copy(struct si_context *sctx,
sctx->dbcb_depth_copy_enabled = true; sctx->dbcb_depth_copy_enabled = true;
if (planes & PIPE_MASK_S) if (planes & PIPE_MASK_S)
sctx->dbcb_stencil_copy_enabled = true; sctx->dbcb_stencil_copy_enabled = true;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
assert(sctx->dbcb_depth_copy_enabled || sctx->dbcb_stencil_copy_enabled); assert(sctx->dbcb_depth_copy_enabled || sctx->dbcb_stencil_copy_enabled);
@ -140,7 +140,7 @@ si_blit_dbcb_copy(struct si_context *sctx,
for (sample = first_sample; sample <= last_sample; sample++) { for (sample = first_sample; sample <= last_sample; sample++) {
if (sample != sctx->dbcb_copy_sample) { if (sample != sctx->dbcb_copy_sample) {
sctx->dbcb_copy_sample = sample; sctx->dbcb_copy_sample = sample;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
} }
si_blitter_begin(sctx, SI_DECOMPRESS); si_blitter_begin(sctx, SI_DECOMPRESS);
@ -161,7 +161,7 @@ si_blit_dbcb_copy(struct si_context *sctx,
sctx->decompression_enabled = false; sctx->decompression_enabled = false;
sctx->dbcb_depth_copy_enabled = false; sctx->dbcb_depth_copy_enabled = false;
sctx->dbcb_stencil_copy_enabled = false; sctx->dbcb_stencil_copy_enabled = false;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
return fully_copied_levels; return fully_copied_levels;
} }
@ -210,7 +210,7 @@ si_blit_decompress_zs_planes_in_place(struct si_context *sctx,
sctx->db_flush_stencil_inplace = true; sctx->db_flush_stencil_inplace = true;
if (planes & PIPE_MASK_Z) if (planes & PIPE_MASK_Z)
sctx->db_flush_depth_inplace = true; sctx->db_flush_depth_inplace = true;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
surf_tmpl.format = texture->resource.b.b.format; surf_tmpl.format = texture->resource.b.b.format;
@ -256,7 +256,7 @@ si_blit_decompress_zs_planes_in_place(struct si_context *sctx,
sctx->decompression_enabled = false; sctx->decompression_enabled = false;
sctx->db_flush_depth_inplace = false; sctx->db_flush_depth_inplace = false;
sctx->db_flush_stencil_inplace = false; sctx->db_flush_stencil_inplace = false;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
} }
/* Helper function of si_flush_depth_texture: decompress the given levels /* Helper function of si_flush_depth_texture: decompress the given levels

14
src/gallium/drivers/radeonsi/si_clear.c
View File

@ -548,7 +548,7 @@ static void si_do_fast_color_clear(struct si_context *sctx,
si_set_clear_color(tex, fb->cbufs[i]->format, color); si_set_clear_color(tex, fb->cbufs[i]->format, color);
sctx->framebuffer.dirty_cbufs |= 1 << i; sctx->framebuffer.dirty_cbufs |= 1 << i;
si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.framebuffer);
*buffers &= ~clear_bit; *buffers &= ~clear_bit;
} }
} }
@ -598,9 +598,9 @@ static void si_clear(struct pipe_context *ctx, unsigned buffers,
zstex->depth_clear_value = depth; zstex->depth_clear_value = depth;
sctx->framebuffer.dirty_zsbuf = true; sctx->framebuffer.dirty_zsbuf = true;
si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_DEPTH_CLEAR */ si_mark_atom_dirty(sctx, &sctx->atoms.s.framebuffer); /* updates DB_DEPTH_CLEAR */
sctx->db_depth_clear = true; sctx->db_depth_clear = true;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
} }
/* TC-compatible HTILE only supports stencil clears to 0. */ /* TC-compatible HTILE only supports stencil clears to 0. */
@ -616,9 +616,9 @@ static void si_clear(struct pipe_context *ctx, unsigned buffers,
zstex->stencil_clear_value = stencil; zstex->stencil_clear_value = stencil;
sctx->framebuffer.dirty_zsbuf = true; sctx->framebuffer.dirty_zsbuf = true;
si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_STENCIL_CLEAR */ si_mark_atom_dirty(sctx, &sctx->atoms.s.framebuffer); /* updates DB_STENCIL_CLEAR */
sctx->db_stencil_clear = true; sctx->db_stencil_clear = true;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
} }
/* TODO: Find out what's wrong here. Fast depth clear leads to /* TODO: Find out what's wrong here. Fast depth clear leads to
@ -647,14 +647,14 @@ static void si_clear(struct pipe_context *ctx, unsigned buffers,
sctx->db_depth_clear = false; sctx->db_depth_clear = false;
sctx->db_depth_disable_expclear = false; sctx->db_depth_disable_expclear = false;
zstex->depth_cleared = true; zstex->depth_cleared = true;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
} }
if (sctx->db_stencil_clear) { if (sctx->db_stencil_clear) {
sctx->db_stencil_clear = false; sctx->db_stencil_clear = false;
sctx->db_stencil_disable_expclear = false; sctx->db_stencil_disable_expclear = false;
zstex->stencil_cleared = true; zstex->stencil_cleared = true;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
} }
} }

8
src/gallium/drivers/radeonsi/si_compute.c
View File

@ -863,10 +863,10 @@ static void si_launch_grid(
si_upload_compute_shader_descriptors(sctx); si_upload_compute_shader_descriptors(sctx);
si_emit_compute_shader_pointers(sctx); si_emit_compute_shader_pointers(sctx);
if (si_is_atom_dirty(sctx, sctx->atoms.s.render_cond)) { if (si_is_atom_dirty(sctx, &sctx->atoms.s.render_cond)) {
sctx->atoms.s.render_cond->emit(sctx, sctx->atoms.s.render_cond.emit(sctx,
sctx->atoms.s.render_cond); &sctx->atoms.s.render_cond);
si_set_atom_dirty(sctx, sctx->atoms.s.render_cond, false); si_set_atom_dirty(sctx, &sctx->atoms.s.render_cond, false);
} }
if ((program->input_size || if ((program->input_size ||

2
src/gallium/drivers/radeonsi/si_cp_dma.c
View File

@ -388,7 +388,7 @@ static void si_cp_dma_realign_engine(struct si_context *sctx, unsigned size,
if (!sctx->scratch_buffer) if (!sctx->scratch_buffer)
return; return;
si_mark_atom_dirty(sctx, &sctx->scratch_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.scratch_state);
} }
si_cp_dma_prepare(sctx, &sctx->scratch_buffer->b.b, si_cp_dma_prepare(sctx, &sctx->scratch_buffer->b.b,

12
src/gallium/drivers/radeonsi/si_descriptors.c
View File

@ -162,7 +162,7 @@ static bool si_upload_descriptors(struct si_context *sctx,
r600_resource_reference(&desc->buffer, NULL); r600_resource_reference(&desc->buffer, NULL);
desc->gpu_list = NULL; desc->gpu_list = NULL;
desc->gpu_address = si_desc_extract_buffer_address(descriptor); desc->gpu_address = si_desc_extract_buffer_address(descriptor);
si_mark_atom_dirty(sctx, &sctx->shader_pointers.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.shader_pointers);
return true; return true;
} }
@ -192,7 +192,7 @@ static bool si_upload_descriptors(struct si_context *sctx,
assert((desc->buffer->gpu_address >> 32) == sctx->screen->info.address32_hi); assert((desc->buffer->gpu_address >> 32) == sctx->screen->info.address32_hi);
assert((desc->gpu_address >> 32) == sctx->screen->info.address32_hi); assert((desc->gpu_address >> 32) == sctx->screen->info.address32_hi);
si_mark_atom_dirty(sctx, &sctx->shader_pointers.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.shader_pointers);
return true; return true;
} }
@ -1172,7 +1172,7 @@ bool si_upload_vertex_buffer_descriptors(struct si_context *sctx)
* on performance (confirmed by testing). New descriptors are always * on performance (confirmed by testing). New descriptors are always
* uploaded to a fresh new buffer, so I don't think flushing the const * uploaded to a fresh new buffer, so I don't think flushing the const
* cache is needed. */ * cache is needed. */
si_mark_atom_dirty(sctx, &sctx->shader_pointers.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.shader_pointers);
sctx->vertex_buffers_dirty = false; sctx->vertex_buffers_dirty = false;
sctx->vertex_buffer_pointer_dirty = true; sctx->vertex_buffer_pointer_dirty = true;
sctx->prefetch_L2_mask |= SI_PREFETCH_VBO_DESCRIPTORS; sctx->prefetch_L2_mask |= SI_PREFETCH_VBO_DESCRIPTORS;
@ -1985,14 +1985,14 @@ static void si_mark_shader_pointers_dirty(struct si_context *sctx,
if (shader == PIPE_SHADER_VERTEX) if (shader == PIPE_SHADER_VERTEX)
sctx->vertex_buffer_pointer_dirty = sctx->vb_descriptors_buffer != NULL; sctx->vertex_buffer_pointer_dirty = sctx->vb_descriptors_buffer != NULL;
si_mark_atom_dirty(sctx, &sctx->shader_pointers.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.shader_pointers);
} }
static void si_shader_pointers_begin_new_cs(struct si_context *sctx) static void si_shader_pointers_begin_new_cs(struct si_context *sctx)
{ {
sctx->shader_pointers_dirty = u_bit_consecutive(0, SI_NUM_DESCS); sctx->shader_pointers_dirty = u_bit_consecutive(0, SI_NUM_DESCS);
sctx->vertex_buffer_pointer_dirty = sctx->vb_descriptors_buffer != NULL; sctx->vertex_buffer_pointer_dirty = sctx->vb_descriptors_buffer != NULL;
si_mark_atom_dirty(sctx, &sctx->shader_pointers.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.shader_pointers);
sctx->graphics_bindless_pointer_dirty = sctx->bindless_descriptors.buffer != NULL; sctx->graphics_bindless_pointer_dirty = sctx->bindless_descriptors.buffer != NULL;
sctx->compute_bindless_pointer_dirty = sctx->bindless_descriptors.buffer != NULL; sctx->compute_bindless_pointer_dirty = sctx->bindless_descriptors.buffer != NULL;
} }
@ -2762,7 +2762,7 @@ void si_init_all_descriptors(struct si_context *sctx)
sctx->b.make_image_handle_resident = si_make_image_handle_resident; sctx->b.make_image_handle_resident = si_make_image_handle_resident;
/* Shader user data. */ /* Shader user data. */
si_init_atom(sctx, &sctx->shader_pointers.atom, &sctx->atoms.s.shader_pointers, si_init_atom(sctx, &sctx->atoms.s.shader_pointers,
si_emit_graphics_shader_pointers); si_emit_graphics_shader_pointers);
/* Set default and immutable mappings. */ /* Set default and immutable mappings. */

34
src/gallium/drivers/radeonsi/si_gfx_cs.c
View File

@ -257,39 +257,39 @@ void si_begin_new_gfx_cs(struct si_context *ctx)
} }
/* This should always be marked as dirty to set the framebuffer scissor /* This should always be marked as dirty to set the framebuffer scissor
* at least. */ * at least. */
si_mark_atom_dirty(ctx, &ctx->framebuffer.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.framebuffer);
si_mark_atom_dirty(ctx, &ctx->clip_regs); si_mark_atom_dirty(ctx, &ctx->atoms.s.clip_regs);
/* CLEAR_STATE sets zeros. */ /* CLEAR_STATE sets zeros. */
if (!has_clear_state || ctx->clip_state.any_nonzeros) if (!has_clear_state || ctx->clip_state.any_nonzeros)
si_mark_atom_dirty(ctx, &ctx->clip_state.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.clip_state);
ctx->msaa_sample_locs.nr_samples = 0; ctx->msaa_sample_locs.nr_samples = 0;
si_mark_atom_dirty(ctx, &ctx->msaa_sample_locs.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.msaa_sample_locs);
si_mark_atom_dirty(ctx, &ctx->msaa_config); si_mark_atom_dirty(ctx, &ctx->atoms.s.msaa_config);
/* CLEAR_STATE sets 0xffff. */ /* CLEAR_STATE sets 0xffff. */
if (!has_clear_state || ctx->sample_mask.sample_mask != 0xffff) if (!has_clear_state || ctx->sample_mask.sample_mask != 0xffff)
si_mark_atom_dirty(ctx, &ctx->sample_mask.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.sample_mask);
si_mark_atom_dirty(ctx, &ctx->cb_render_state); si_mark_atom_dirty(ctx, &ctx->atoms.s.cb_render_state);
/* CLEAR_STATE sets zeros. */ /* CLEAR_STATE sets zeros. */
if (!has_clear_state || ctx->blend_color.any_nonzeros) if (!has_clear_state || ctx->blend_color.any_nonzeros)
si_mark_atom_dirty(ctx, &ctx->blend_color.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.blend_color);
si_mark_atom_dirty(ctx, &ctx->db_render_state); si_mark_atom_dirty(ctx, &ctx->atoms.s.db_render_state);
if (ctx->chip_class >= GFX9) if (ctx->chip_class >= GFX9)
si_mark_atom_dirty(ctx, &ctx->dpbb_state); si_mark_atom_dirty(ctx, &ctx->atoms.s.dpbb_state);
si_mark_atom_dirty(ctx, &ctx->stencil_ref.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.stencil_ref);
si_mark_atom_dirty(ctx, &ctx->spi_map); si_mark_atom_dirty(ctx, &ctx->atoms.s.spi_map);
si_mark_atom_dirty(ctx, &ctx->streamout.enable_atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.streamout_enable);
si_mark_atom_dirty(ctx, &ctx->render_cond_atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.render_cond);
si_all_descriptors_begin_new_cs(ctx); si_all_descriptors_begin_new_cs(ctx);
si_all_resident_buffers_begin_new_cs(ctx); si_all_resident_buffers_begin_new_cs(ctx);
ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
ctx->viewports.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; ctx->viewports.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
si_mark_atom_dirty(ctx, &ctx->scissors.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
si_mark_atom_dirty(ctx, &ctx->viewports.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
si_mark_atom_dirty(ctx, &ctx->scratch_state); si_mark_atom_dirty(ctx, &ctx->atoms.s.scratch_state);
if (ctx->scratch_buffer) { if (ctx->scratch_buffer) {
si_context_add_resource_size(ctx, &ctx->scratch_buffer->b.b); si_context_add_resource_size(ctx, &ctx->scratch_buffer->b.b);
} }

17
src/gallium/drivers/radeonsi/si_pipe.h
View File

@ -539,7 +539,6 @@ struct si_screen {
}; };
struct si_blend_color { struct si_blend_color {
struct si_atom atom;
struct pipe_blend_color state; struct pipe_blend_color state;
bool any_nonzeros; bool any_nonzeros;
}; };
@ -594,7 +593,6 @@ struct si_images {
}; };
struct si_framebuffer { struct si_framebuffer {
struct si_atom atom;
struct pipe_framebuffer_state state; struct pipe_framebuffer_state state;
unsigned colorbuf_enabled_4bit; unsigned colorbuf_enabled_4bit;
unsigned spi_shader_col_format; unsigned spi_shader_col_format;
@ -622,13 +620,11 @@ struct si_signed_scissor {
}; };
struct si_scissors { struct si_scissors {
struct si_atom atom;
unsigned dirty_mask; unsigned dirty_mask;
struct pipe_scissor_state states[SI_MAX_VIEWPORTS]; struct pipe_scissor_state states[SI_MAX_VIEWPORTS];
}; };
struct si_viewports { struct si_viewports {
struct si_atom atom;
unsigned dirty_mask; unsigned dirty_mask;
unsigned depth_range_dirty_mask; unsigned depth_range_dirty_mask;
struct pipe_viewport_state states[SI_MAX_VIEWPORTS]; struct pipe_viewport_state states[SI_MAX_VIEWPORTS];
@ -636,18 +632,15 @@ struct si_viewports {
}; };
struct si_clip_state { struct si_clip_state {
struct si_atom atom;
struct pipe_clip_state state; struct pipe_clip_state state;
bool any_nonzeros; bool any_nonzeros;
}; };
struct si_sample_locs { struct si_sample_locs {
struct si_atom atom;
unsigned nr_samples; unsigned nr_samples;
}; };
struct si_sample_mask { struct si_sample_mask {
struct si_atom atom;
uint16_t sample_mask; uint16_t sample_mask;
}; };
@ -663,7 +656,6 @@ struct si_streamout_target {
}; };
struct si_streamout { struct si_streamout {
struct si_atom begin_atom;
bool begin_emitted; bool begin_emitted;
unsigned enabled_mask; unsigned enabled_mask;
@ -682,7 +674,6 @@ struct si_streamout {
unsigned hw_enabled_mask; unsigned hw_enabled_mask;
/* The state of VGT_STRMOUT_(CONFIG|EN). */ /* The state of VGT_STRMOUT_(CONFIG|EN). */
struct si_atom enable_atom;
bool streamout_enabled; bool streamout_enabled;
bool prims_gen_query_enabled; bool prims_gen_query_enabled;
int num_prims_gen_queries; int num_prims_gen_queries;
@ -825,18 +816,12 @@ struct si_context {
/* Atom declarations. */ /* Atom declarations. */
struct si_framebuffer framebuffer; struct si_framebuffer framebuffer;
struct si_sample_locs msaa_sample_locs; struct si_sample_locs msaa_sample_locs;
struct si_atom db_render_state;
struct si_atom dpbb_state;
struct si_atom msaa_config;
struct si_sample_mask sample_mask; struct si_sample_mask sample_mask;
struct si_atom cb_render_state;
unsigned last_cb_target_mask; unsigned last_cb_target_mask;
struct si_blend_color blend_color; struct si_blend_color blend_color;
struct si_atom clip_regs;
struct si_clip_state clip_state; struct si_clip_state clip_state;
struct si_shader_data shader_pointers; struct si_shader_data shader_pointers;
struct si_stencil_ref stencil_ref; struct si_stencil_ref stencil_ref;
struct si_atom spi_map;
struct si_scissors scissors; struct si_scissors scissors;
struct si_streamout streamout; struct si_streamout streamout;
struct si_viewports viewports; struct si_viewports viewports;
@ -932,7 +917,6 @@ struct si_context {
enum pipe_prim_type current_rast_prim; /* primitive type after TES, GS */ enum pipe_prim_type current_rast_prim; /* primitive type after TES, GS */
/* Scratch buffer */ /* Scratch buffer */
struct si_atom scratch_state;
struct r600_resource *scratch_buffer; struct r600_resource *scratch_buffer;
unsigned scratch_waves; unsigned scratch_waves;
unsigned spi_tmpring_size; unsigned spi_tmpring_size;
@ -1028,7 +1012,6 @@ struct si_context {
unsigned num_cs_dw_queries_suspend; unsigned num_cs_dw_queries_suspend;
/* Render condition. */ /* Render condition. */
struct si_atom render_cond_atom;
struct pipe_query *render_cond; struct pipe_query *render_cond;
unsigned render_cond_mode; unsigned render_cond_mode;
bool render_cond_invert; bool render_cond_invert;

4
src/gallium/drivers/radeonsi/si_query.c
View File

@ -1776,7 +1776,7 @@ static void si_render_condition(struct pipe_context *ctx,
{ {
struct si_context *sctx = (struct si_context *)ctx; struct si_context *sctx = (struct si_context *)ctx;
struct si_query_hw *rquery = (struct si_query_hw *)query; struct si_query_hw *rquery = (struct si_query_hw *)query;
struct si_atom *atom = &sctx->render_cond_atom; struct si_atom *atom = &sctx->atoms.s.render_cond;
if (query) { if (query) {
bool needs_workaround = false; bool needs_workaround = false;
@ -2053,7 +2053,7 @@ void si_init_query_functions(struct si_context *sctx)
sctx->b.end_query = si_end_query; sctx->b.end_query = si_end_query;
sctx->b.get_query_result = si_get_query_result; sctx->b.get_query_result = si_get_query_result;
sctx->b.get_query_result_resource = si_get_query_result_resource; sctx->b.get_query_result_resource = si_get_query_result_resource;
sctx->render_cond_atom.emit = si_emit_query_predication; sctx->atoms.s.render_cond.emit = si_emit_query_predication;
if (((struct si_screen*)sctx->b.screen)->info.num_render_backends > 0) if (((struct si_screen*)sctx->b.screen)->info.num_render_backends > 0)
sctx->b.render_condition = si_render_condition; sctx->b.render_condition = si_render_condition;

98
src/gallium/drivers/radeonsi/si_state.c
View File

@ -35,21 +35,17 @@
/* Initialize an external atom (owned by ../radeon). */ /* Initialize an external atom (owned by ../radeon). */
static void static void
si_init_external_atom(struct si_context *sctx, struct si_atom *atom, si_init_external_atom(struct si_context *sctx, struct si_atom *atom)
struct si_atom **list_elem)
{ {
atom->id = list_elem - sctx->atoms.array; atom->id = atom - sctx->atoms.array;
*list_elem = atom;
} }
/* Initialize an atom owned by radeonsi. */ /* Initialize an atom owned by radeonsi. */
void si_init_atom(struct si_context *sctx, struct si_atom *atom, void si_init_atom(struct si_context *sctx, struct si_atom *atom,
struct si_atom **list_elem,
void (*emit_func)(struct si_context *ctx, struct si_atom *state)) void (*emit_func)(struct si_context *ctx, struct si_atom *state))
{ {
atom->emit = emit_func; atom->emit = emit_func;
atom->id = list_elem - sctx->atoms.array; atom->id = atom - sctx->atoms.array;
*list_elem = atom;
} }
static unsigned si_map_swizzle(unsigned swizzle) static unsigned si_map_swizzle(unsigned swizzle)
@ -672,7 +668,7 @@ static void si_bind_blend_state(struct pipe_context *ctx, void *state)
(old_blend->blend_enable_4bit != blend->blend_enable_4bit && (old_blend->blend_enable_4bit != blend->blend_enable_4bit &&
sctx->framebuffer.nr_samples >= 2 && sctx->framebuffer.nr_samples >= 2 &&
sctx->screen->dcc_msaa_allowed)) sctx->screen->dcc_msaa_allowed))
si_mark_atom_dirty(sctx, &sctx->cb_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.cb_render_state);
if (!old_blend || if (!old_blend ||
old_blend->cb_target_mask != blend->cb_target_mask || old_blend->cb_target_mask != blend->cb_target_mask ||
@ -688,7 +684,7 @@ static void si_bind_blend_state(struct pipe_context *ctx, void *state)
old_blend->alpha_to_coverage != blend->alpha_to_coverage || old_blend->alpha_to_coverage != blend->alpha_to_coverage ||
old_blend->blend_enable_4bit != blend->blend_enable_4bit || old_blend->blend_enable_4bit != blend->blend_enable_4bit ||
old_blend->cb_target_enabled_4bit != blend->cb_target_enabled_4bit)) old_blend->cb_target_enabled_4bit != blend->cb_target_enabled_4bit))
si_mark_atom_dirty(sctx, &sctx->dpbb_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.dpbb_state);
if (sctx->screen->has_out_of_order_rast && if (sctx->screen->has_out_of_order_rast &&
(!old_blend || (!old_blend ||
@ -696,7 +692,7 @@ static void si_bind_blend_state(struct pipe_context *ctx, void *state)
old_blend->cb_target_enabled_4bit != blend->cb_target_enabled_4bit || old_blend->cb_target_enabled_4bit != blend->cb_target_enabled_4bit ||
old_blend->commutative_4bit != blend->commutative_4bit || old_blend->commutative_4bit != blend->commutative_4bit ||
old_blend->logicop_enable != blend->logicop_enable))) old_blend->logicop_enable != blend->logicop_enable)))
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
} }
static void si_delete_blend_state(struct pipe_context *ctx, void *state) static void si_delete_blend_state(struct pipe_context *ctx, void *state)
@ -713,7 +709,7 @@ static void si_set_blend_color(struct pipe_context *ctx,
sctx->blend_color.state = *state; sctx->blend_color.state = *state;
sctx->blend_color.any_nonzeros = memcmp(state, &zeros, sizeof(*state)) != 0; sctx->blend_color.any_nonzeros = memcmp(state, &zeros, sizeof(*state)) != 0;
si_mark_atom_dirty(sctx, &sctx->blend_color.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.blend_color);
} }
static void si_emit_blend_color(struct si_context *sctx, struct si_atom *atom) static void si_emit_blend_color(struct si_context *sctx, struct si_atom *atom)
@ -740,7 +736,7 @@ static void si_set_clip_state(struct pipe_context *ctx,
sctx->clip_state.state = *state; sctx->clip_state.state = *state;
sctx->clip_state.any_nonzeros = memcmp(state, &zeros, sizeof(*state)) != 0; sctx->clip_state.any_nonzeros = memcmp(state, &zeros, sizeof(*state)) != 0;
si_mark_atom_dirty(sctx, &sctx->clip_state.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.clip_state);
cb.buffer = NULL; cb.buffer = NULL;
cb.user_buffer = state->ucp; cb.user_buffer = state->ucp;
@ -1006,12 +1002,12 @@ static void si_bind_rs_state(struct pipe_context *ctx, void *state)
return; return;
if (!old_rs || old_rs->multisample_enable != rs->multisample_enable) { if (!old_rs || old_rs->multisample_enable != rs->multisample_enable) {
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
/* Update the small primitive filter workaround if necessary. */ /* Update the small primitive filter workaround if necessary. */
if (sctx->screen->has_msaa_sample_loc_bug && if (sctx->screen->has_msaa_sample_loc_bug &&
sctx->framebuffer.nr_samples > 1) sctx->framebuffer.nr_samples > 1)
si_mark_atom_dirty(sctx, &sctx->msaa_sample_locs.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_sample_locs);
} }
sctx->current_vs_state &= C_VS_STATE_CLAMP_VERTEX_COLOR; sctx->current_vs_state &= C_VS_STATE_CLAMP_VERTEX_COLOR;
@ -1025,19 +1021,19 @@ static void si_bind_rs_state(struct pipe_context *ctx, void *state)
old_rs->line_width != rs->line_width || old_rs->line_width != rs->line_width ||
old_rs->max_point_size != rs->max_point_size)) { old_rs->max_point_size != rs->max_point_size)) {
sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
si_mark_atom_dirty(sctx, &sctx->scissors.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.scissors);
} }
if (!old_rs || if (!old_rs ||
old_rs->clip_halfz != rs->clip_halfz) { old_rs->clip_halfz != rs->clip_halfz) {
sctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; sctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
si_mark_atom_dirty(sctx, &sctx->viewports.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.viewports);
} }
if (!old_rs || if (!old_rs ||
old_rs->clip_plane_enable != rs->clip_plane_enable || old_rs->clip_plane_enable != rs->clip_plane_enable ||
old_rs->pa_cl_clip_cntl != rs->pa_cl_clip_cntl) old_rs->pa_cl_clip_cntl != rs->pa_cl_clip_cntl)
si_mark_atom_dirty(sctx, &sctx->clip_regs); si_mark_atom_dirty(sctx, &sctx->atoms.s.clip_regs);
sctx->ia_multi_vgt_param_key.u.line_stipple_enabled = sctx->ia_multi_vgt_param_key.u.line_stipple_enabled =
rs->line_stipple_enable; rs->line_stipple_enable;
@ -1098,7 +1094,7 @@ static void si_set_stencil_ref(struct pipe_context *ctx,
return; return;
sctx->stencil_ref.state = *state; sctx->stencil_ref.state = *state;
si_mark_atom_dirty(sctx, &sctx->stencil_ref.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.stencil_ref);
} }
@ -1286,7 +1282,7 @@ static void si_bind_dsa_state(struct pipe_context *ctx, void *state)
if (memcmp(&dsa->stencil_ref, &sctx->stencil_ref.dsa_part, if (memcmp(&dsa->stencil_ref, &sctx->stencil_ref.dsa_part,
sizeof(struct si_dsa_stencil_ref_part)) != 0) { sizeof(struct si_dsa_stencil_ref_part)) != 0) {
sctx->stencil_ref.dsa_part = dsa->stencil_ref; sctx->stencil_ref.dsa_part = dsa->stencil_ref;
si_mark_atom_dirty(sctx, &sctx->stencil_ref.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.stencil_ref);
} }
if (!old_dsa || old_dsa->alpha_func != dsa->alpha_func) if (!old_dsa || old_dsa->alpha_func != dsa->alpha_func)
@ -1297,13 +1293,13 @@ static void si_bind_dsa_state(struct pipe_context *ctx, void *state)
(old_dsa->depth_enabled != dsa->depth_enabled || (old_dsa->depth_enabled != dsa->depth_enabled ||
old_dsa->stencil_enabled != dsa->stencil_enabled || old_dsa->stencil_enabled != dsa->stencil_enabled ||
old_dsa->db_can_write != dsa->db_can_write))) old_dsa->db_can_write != dsa->db_can_write)))
si_mark_atom_dirty(sctx, &sctx->dpbb_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.dpbb_state);
if (sctx->screen->has_out_of_order_rast && if (sctx->screen->has_out_of_order_rast &&
(!old_dsa || (!old_dsa ||
memcmp(old_dsa->order_invariance, dsa->order_invariance, memcmp(old_dsa->order_invariance, dsa->order_invariance,
sizeof(old_dsa->order_invariance)))) sizeof(old_dsa->order_invariance))))
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
} }
static void si_delete_dsa_state(struct pipe_context *ctx, void *state) static void si_delete_dsa_state(struct pipe_context *ctx, void *state)
@ -1337,19 +1333,19 @@ static void si_set_active_query_state(struct pipe_context *ctx, boolean enable)
/* Occlusion queries. */ /* Occlusion queries. */
if (sctx->occlusion_queries_disabled != !enable) { if (sctx->occlusion_queries_disabled != !enable) {
sctx->occlusion_queries_disabled = !enable; sctx->occlusion_queries_disabled = !enable;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
} }
} }
void si_set_occlusion_query_state(struct si_context *sctx, void si_set_occlusion_query_state(struct si_context *sctx,
bool old_perfect_enable) bool old_perfect_enable)
{ {
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
bool perfect_enable = sctx->num_perfect_occlusion_queries != 0; bool perfect_enable = sctx->num_perfect_occlusion_queries != 0;
if (perfect_enable != old_perfect_enable) if (perfect_enable != old_perfect_enable)
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
} }
void si_save_qbo_state(struct si_context *sctx, struct si_qbo_state *st) void si_save_qbo_state(struct si_context *sctx, struct si_qbo_state *st)
@ -2895,24 +2891,24 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
si_update_ps_colorbuf0_slot(sctx); si_update_ps_colorbuf0_slot(sctx);
si_update_poly_offset_state(sctx); si_update_poly_offset_state(sctx);
si_mark_atom_dirty(sctx, &sctx->cb_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.cb_render_state);
si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.framebuffer);
if (sctx->screen->dpbb_allowed) if (sctx->screen->dpbb_allowed)
si_mark_atom_dirty(sctx, &sctx->dpbb_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.dpbb_state);
if (sctx->framebuffer.any_dst_linear != old_any_dst_linear) if (sctx->framebuffer.any_dst_linear != old_any_dst_linear)
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
if (sctx->screen->has_out_of_order_rast && if (sctx->screen->has_out_of_order_rast &&
(sctx->framebuffer.colorbuf_enabled_4bit != old_colorbuf_enabled_4bit || (sctx->framebuffer.colorbuf_enabled_4bit != old_colorbuf_enabled_4bit ||
!!sctx->framebuffer.state.zsbuf != old_has_zsbuf || !!sctx->framebuffer.state.zsbuf != old_has_zsbuf ||
(zstex && zstex->surface.has_stencil != old_has_stencil))) (zstex && zstex->surface.has_stencil != old_has_stencil)))
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
if (sctx->framebuffer.nr_samples != old_nr_samples) { if (sctx->framebuffer.nr_samples != old_nr_samples) {
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
/* Set sample locations as fragment shader constants. */ /* Set sample locations as fragment shader constants. */
switch (sctx->framebuffer.nr_samples) { switch (sctx->framebuffer.nr_samples) {
@ -2939,7 +2935,7 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
constbuf.buffer_size = sctx->framebuffer.nr_samples * 2 * 4; constbuf.buffer_size = sctx->framebuffer.nr_samples * 2 * 4;
si_set_rw_buffer(sctx, SI_PS_CONST_SAMPLE_POSITIONS, &constbuf); si_set_rw_buffer(sctx, SI_PS_CONST_SAMPLE_POSITIONS, &constbuf);
si_mark_atom_dirty(sctx, &sctx->msaa_sample_locs.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_sample_locs);
} }
sctx->do_update_shaders = true; sctx->do_update_shaders = true;
@ -3416,9 +3412,9 @@ static void si_emit_msaa_config(struct si_context *sctx, struct si_atom *atom)
void si_update_ps_iter_samples(struct si_context *sctx) void si_update_ps_iter_samples(struct si_context *sctx)
{ {
if (sctx->framebuffer.nr_samples > 1) if (sctx->framebuffer.nr_samples > 1)
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
if (sctx->screen->dpbb_allowed) if (sctx->screen->dpbb_allowed)
si_mark_atom_dirty(sctx, &sctx->dpbb_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.dpbb_state);
} }
static void si_set_min_samples(struct pipe_context *ctx, unsigned min_samples) static void si_set_min_samples(struct pipe_context *ctx, unsigned min_samples)
@ -4174,7 +4170,7 @@ static void si_set_sample_mask(struct pipe_context *ctx, unsigned sample_mask)
return; return;
sctx->sample_mask.sample_mask = sample_mask; sctx->sample_mask.sample_mask = sample_mask;
si_mark_atom_dirty(sctx, &sctx->sample_mask.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.sample_mask);
} }
static void si_emit_sample_mask(struct si_context *sctx, struct si_atom *atom) static void si_emit_sample_mask(struct si_context *sctx, struct si_atom *atom)
@ -4530,23 +4526,23 @@ static void si_init_config(struct si_context *sctx);
void si_init_state_functions(struct si_context *sctx) void si_init_state_functions(struct si_context *sctx)
{ {
si_init_external_atom(sctx, &sctx->render_cond_atom, &sctx->atoms.s.render_cond); si_init_external_atom(sctx, &sctx->atoms.s.render_cond);
si_init_external_atom(sctx, &sctx->streamout.begin_atom, &sctx->atoms.s.streamout_begin); si_init_external_atom(sctx, &sctx->atoms.s.streamout_begin);
si_init_external_atom(sctx, &sctx->streamout.enable_atom, &sctx->atoms.s.streamout_enable); si_init_external_atom(sctx, &sctx->atoms.s.streamout_enable);
si_init_external_atom(sctx, &sctx->scissors.atom, &sctx->atoms.s.scissors); si_init_external_atom(sctx, &sctx->atoms.s.scissors);
si_init_external_atom(sctx, &sctx->viewports.atom, &sctx->atoms.s.viewports); si_init_external_atom(sctx, &sctx->atoms.s.viewports);
si_init_atom(sctx, &sctx->framebuffer.atom, &sctx->atoms.s.framebuffer, si_emit_framebuffer_state); si_init_atom(sctx, &sctx->atoms.s.framebuffer, si_emit_framebuffer_state);
si_init_atom(sctx, &sctx->msaa_sample_locs.atom, &sctx->atoms.s.msaa_sample_locs, si_emit_msaa_sample_locs); si_init_atom(sctx, &sctx->atoms.s.msaa_sample_locs, si_emit_msaa_sample_locs);
si_init_atom(sctx, &sctx->db_render_state, &sctx->atoms.s.db_render_state, si_emit_db_render_state); si_init_atom(sctx, &sctx->atoms.s.db_render_state, si_emit_db_render_state);
si_init_atom(sctx, &sctx->dpbb_state, &sctx->atoms.s.dpbb_state, si_emit_dpbb_state); si_init_atom(sctx, &sctx->atoms.s.dpbb_state, si_emit_dpbb_state);
si_init_atom(sctx, &sctx->msaa_config, &sctx->atoms.s.msaa_config, si_emit_msaa_config); si_init_atom(sctx, &sctx->atoms.s.msaa_config, si_emit_msaa_config);
si_init_atom(sctx, &sctx->sample_mask.atom, &sctx->atoms.s.sample_mask, si_emit_sample_mask); si_init_atom(sctx, &sctx->atoms.s.sample_mask, si_emit_sample_mask);
si_init_atom(sctx, &sctx->cb_render_state, &sctx->atoms.s.cb_render_state, si_emit_cb_render_state); si_init_atom(sctx, &sctx->atoms.s.cb_render_state, si_emit_cb_render_state);
si_init_atom(sctx, &sctx->blend_color.atom, &sctx->atoms.s.blend_color, si_emit_blend_color); si_init_atom(sctx, &sctx->atoms.s.blend_color, si_emit_blend_color);
si_init_atom(sctx, &sctx->clip_regs, &sctx->atoms.s.clip_regs, si_emit_clip_regs); si_init_atom(sctx, &sctx->atoms.s.clip_regs, si_emit_clip_regs);
si_init_atom(sctx, &sctx->clip_state.atom, &sctx->atoms.s.clip_state, si_emit_clip_state); si_init_atom(sctx, &sctx->atoms.s.clip_state, si_emit_clip_state);
si_init_atom(sctx, &sctx->stencil_ref.atom, &sctx->atoms.s.stencil_ref, si_emit_stencil_ref); si_init_atom(sctx, &sctx->atoms.s.stencil_ref, si_emit_stencil_ref);
sctx->b.create_blend_state = si_create_blend_state; sctx->b.create_blend_state = si_create_blend_state;
sctx->b.bind_blend_state = si_bind_blend_state; sctx->b.bind_blend_state = si_bind_blend_state;

43
src/gallium/drivers/radeonsi/si_state.h
View File

@ -132,7 +132,6 @@ struct si_state_dsa {
}; };
struct si_stencil_ref { struct si_stencil_ref {
struct si_atom atom;
struct pipe_stencil_ref state; struct pipe_stencil_ref state;
struct si_dsa_stencil_ref_part dsa_part; struct si_dsa_stencil_ref_part dsa_part;
}; };
@ -178,33 +177,32 @@ union si_state {
union si_state_atoms { union si_state_atoms {
struct { struct {
/* The order matters. */ /* The order matters. */
struct si_atom *render_cond; struct si_atom render_cond;
struct si_atom *streamout_begin; struct si_atom streamout_begin;
struct si_atom *streamout_enable; /* must be after streamout_begin */ struct si_atom streamout_enable; /* must be after streamout_begin */
struct si_atom *framebuffer; struct si_atom framebuffer;
struct si_atom *msaa_sample_locs; struct si_atom msaa_sample_locs;
struct si_atom *db_render_state; struct si_atom db_render_state;
struct si_atom *dpbb_state; struct si_atom dpbb_state;
struct si_atom *msaa_config; struct si_atom msaa_config;
struct si_atom *sample_mask; struct si_atom sample_mask;
struct si_atom *cb_render_state; struct si_atom cb_render_state;
struct si_atom *blend_color; struct si_atom blend_color;
struct si_atom *clip_regs; struct si_atom clip_regs;
struct si_atom *clip_state; struct si_atom clip_state;
struct si_atom *shader_pointers; struct si_atom shader_pointers;
struct si_atom *scissors; struct si_atom scissors;
struct si_atom *viewports; struct si_atom viewports;
struct si_atom *stencil_ref; struct si_atom stencil_ref;
struct si_atom *spi_map; struct si_atom spi_map;
struct si_atom *scratch_state; struct si_atom scratch_state;
} s; } s;
struct si_atom *array[0]; struct si_atom array[0];
}; };
#define SI_NUM_ATOMS (sizeof(union si_state_atoms)/sizeof(struct si_atom*)) #define SI_NUM_ATOMS (sizeof(union si_state_atoms)/sizeof(struct si_atom*))
struct si_shader_data { struct si_shader_data {
struct si_atom atom;
uint32_t sh_base[SI_NUM_SHADERS]; uint32_t sh_base[SI_NUM_SHADERS];
}; };
@ -384,7 +382,6 @@ void si_rebind_buffer(struct si_context *sctx, struct pipe_resource *buf,
struct si_shader_selector; struct si_shader_selector;
void si_init_atom(struct si_context *sctx, struct si_atom *atom, void si_init_atom(struct si_context *sctx, struct si_atom *atom,
struct si_atom **list_elem,
void (*emit_func)(struct si_context *ctx, struct si_atom *state)); void (*emit_func)(struct si_context *ctx, struct si_atom *state));
void si_init_state_functions(struct si_context *sctx); void si_init_state_functions(struct si_context *sctx);
void si_init_screen_state_functions(struct si_screen *sscreen); void si_init_screen_state_functions(struct si_screen *sscreen);

14
src/gallium/drivers/radeonsi/si_state_draw.c
View File

@ -1182,7 +1182,7 @@ static void si_emit_all_states(struct si_context *sctx, const struct pipe_draw_i
/* Emit state atoms. */ /* Emit state atoms. */
unsigned mask = sctx->dirty_atoms & ~skip_atom_mask; unsigned mask = sctx->dirty_atoms & ~skip_atom_mask;
while (mask) { while (mask) {
struct si_atom *atom = sctx->atoms.array[u_bit_scan(&mask)]; struct si_atom *atom = &sctx->atoms.array[u_bit_scan(&mask)];
atom->emit(sctx, atom); atom->emit(sctx, atom);
} }
@ -1256,7 +1256,7 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
sctx->framebuffer.dirty_cbufs |= sctx->framebuffer.dirty_cbufs |=
((1 << sctx->framebuffer.state.nr_cbufs) - 1); ((1 << sctx->framebuffer.state.nr_cbufs) - 1);
sctx->framebuffer.dirty_zsbuf = true; sctx->framebuffer.dirty_zsbuf = true;
si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.framebuffer);
si_update_all_texture_descriptors(sctx); si_update_all_texture_descriptors(sctx);
} }
@ -1282,7 +1282,7 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
bool new_is_poly = rast_prim >= PIPE_PRIM_TRIANGLES; bool new_is_poly = rast_prim >= PIPE_PRIM_TRIANGLES;
if (old_is_poly != new_is_poly) { if (old_is_poly != new_is_poly) {
sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
si_mark_atom_dirty(sctx, &sctx->scissors.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.scissors);
} }
sctx->current_rast_prim = rast_prim; sctx->current_rast_prim = rast_prim;
@ -1418,7 +1418,7 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
* more involved alternative workaround. * more involved alternative workaround.
*/ */
if ((sctx->family == CHIP_VEGA10 || sctx->family == CHIP_RAVEN) && if ((sctx->family == CHIP_VEGA10 || sctx->family == CHIP_RAVEN) &&
si_is_atom_dirty(sctx, &sctx->scissors.atom)) { si_is_atom_dirty(sctx, &sctx->atoms.s.scissors)) {
sctx->flags |= SI_CONTEXT_PS_PARTIAL_FLUSH; sctx->flags |= SI_CONTEXT_PS_PARTIAL_FLUSH;
si_emit_cache_flush(sctx); si_emit_cache_flush(sctx);
} }
@ -1436,7 +1436,7 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
unsigned masked_atoms = 0; unsigned masked_atoms = 0;
if (unlikely(sctx->flags & SI_CONTEXT_FLUSH_FOR_RENDER_COND)) if (unlikely(sctx->flags & SI_CONTEXT_FLUSH_FOR_RENDER_COND))
masked_atoms |= 1u << sctx->render_cond_atom.id; masked_atoms |= 1u << sctx->atoms.s.render_cond.id;
if (!si_upload_graphics_shader_descriptors(sctx)) if (!si_upload_graphics_shader_descriptors(sctx))
return; return;
@ -1446,8 +1446,8 @@ void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
si_emit_cache_flush(sctx); si_emit_cache_flush(sctx);
/* <-- CUs are idle here. */ /* <-- CUs are idle here. */
if (si_is_atom_dirty(sctx, &sctx->render_cond_atom)) if (si_is_atom_dirty(sctx, &sctx->atoms.s.render_cond))
sctx->render_cond_atom.emit(sctx, NULL); sctx->atoms.s.render_cond.emit(sctx, NULL);
sctx->dirty_atoms = 0; sctx->dirty_atoms = 0;
si_emit_draw_packets(sctx, info, indexbuf, index_size, index_offset); si_emit_draw_packets(sctx, info, indexbuf, index_size, index_offset);

30
src/gallium/drivers/radeonsi/si_state_shaders.c
View File

@ -2272,7 +2272,7 @@ static void si_update_clip_regs(struct si_context *sctx,
!next_hw_vs_variant || !next_hw_vs_variant ||
old_hw_vs_variant->key.opt.clip_disable != old_hw_vs_variant->key.opt.clip_disable !=
next_hw_vs_variant->key.opt.clip_disable)) next_hw_vs_variant->key.opt.clip_disable))
si_mark_atom_dirty(sctx, &sctx->clip_regs); si_mark_atom_dirty(sctx, &sctx->atoms.s.clip_regs);
} }
static void si_update_common_shader_state(struct si_context *sctx) static void si_update_common_shader_state(struct si_context *sctx)
@ -2428,14 +2428,14 @@ static void si_bind_ps_shader(struct pipe_context *ctx, void *state)
if (!old_sel || if (!old_sel ||
old_sel->info.colors_written != sel->info.colors_written) old_sel->info.colors_written != sel->info.colors_written)
si_mark_atom_dirty(sctx, &sctx->cb_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.cb_render_state);
if (sctx->screen->has_out_of_order_rast && if (sctx->screen->has_out_of_order_rast &&
(!old_sel || (!old_sel ||
old_sel->info.writes_memory != sel->info.writes_memory || old_sel->info.writes_memory != sel->info.writes_memory ||
old_sel->info.properties[TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL] != old_sel->info.properties[TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL] !=
sel->info.properties[TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL])) sel->info.properties[TGSI_PROPERTY_FS_EARLY_DEPTH_STENCIL]))
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
} }
si_set_active_descriptors_for_shader(sctx, sel); si_set_active_descriptors_for_shader(sctx, sel);
si_update_ps_colorbuf0_slot(sctx); si_update_ps_colorbuf0_slot(sctx);
@ -2980,7 +2980,7 @@ static bool si_update_spi_tmpring_size(struct si_context *sctx)
if (!sctx->scratch_buffer) if (!sctx->scratch_buffer)
return false; return false;
si_mark_atom_dirty(sctx, &sctx->scratch_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.scratch_state);
si_context_add_resource_size(sctx, si_context_add_resource_size(sctx,
&sctx->scratch_buffer->b.b); &sctx->scratch_buffer->b.b);
} }
@ -2997,7 +2997,7 @@ static bool si_update_spi_tmpring_size(struct si_context *sctx)
S_0286E8_WAVESIZE(scratch_bytes_per_wave >> 10); S_0286E8_WAVESIZE(scratch_bytes_per_wave >> 10);
if (spi_tmpring_size != sctx->spi_tmpring_size) { if (spi_tmpring_size != sctx->spi_tmpring_size) {
sctx->spi_tmpring_size = spi_tmpring_size; sctx->spi_tmpring_size = spi_tmpring_size;
si_mark_atom_dirty(sctx, &sctx->scratch_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.scratch_state);
} }
return true; return true;
} }
@ -3234,7 +3234,7 @@ bool si_update_shaders(struct si_context *sctx)
si_update_vgt_shader_config(sctx); si_update_vgt_shader_config(sctx);
if (old_clip_disable != si_get_vs_state(sctx)->key.opt.clip_disable) if (old_clip_disable != si_get_vs_state(sctx)->key.opt.clip_disable)
si_mark_atom_dirty(sctx, &sctx->clip_regs); si_mark_atom_dirty(sctx, &sctx->atoms.s.clip_regs);
if (sctx->ps_shader.cso) { if (sctx->ps_shader.cso) {
unsigned db_shader_control; unsigned db_shader_control;
@ -3253,7 +3253,7 @@ bool si_update_shaders(struct si_context *sctx)
sctx->flatshade != rs->flatshade) { sctx->flatshade != rs->flatshade) {
sctx->sprite_coord_enable = rs->sprite_coord_enable; sctx->sprite_coord_enable = rs->sprite_coord_enable;
sctx->flatshade = rs->flatshade; sctx->flatshade = rs->flatshade;
si_mark_atom_dirty(sctx, &sctx->spi_map); si_mark_atom_dirty(sctx, &sctx->atoms.s.spi_map);
} }
if (sctx->screen->rbplus_allowed && if (sctx->screen->rbplus_allowed &&
@ -3261,24 +3261,24 @@ bool si_update_shaders(struct si_context *sctx)
(!old_ps || (!old_ps ||
old_spi_shader_col_format != old_spi_shader_col_format !=
sctx->ps_shader.current->key.part.ps.epilog.spi_shader_col_format)) sctx->ps_shader.current->key.part.ps.epilog.spi_shader_col_format))
si_mark_atom_dirty(sctx, &sctx->cb_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.cb_render_state);
if (sctx->ps_db_shader_control != db_shader_control) { if (sctx->ps_db_shader_control != db_shader_control) {
sctx->ps_db_shader_control = db_shader_control; sctx->ps_db_shader_control = db_shader_control;
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
if (sctx->screen->dpbb_allowed) if (sctx->screen->dpbb_allowed)
si_mark_atom_dirty(sctx, &sctx->dpbb_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.dpbb_state);
} }
if (sctx->smoothing_enabled != sctx->ps_shader.current->key.part.ps.epilog.poly_line_smoothing) { if (sctx->smoothing_enabled != sctx->ps_shader.current->key.part.ps.epilog.poly_line_smoothing) {
sctx->smoothing_enabled = sctx->ps_shader.current->key.part.ps.epilog.poly_line_smoothing; sctx->smoothing_enabled = sctx->ps_shader.current->key.part.ps.epilog.poly_line_smoothing;
si_mark_atom_dirty(sctx, &sctx->msaa_config); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
if (sctx->chip_class == SI) if (sctx->chip_class == SI)
si_mark_atom_dirty(sctx, &sctx->db_render_state); si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
if (sctx->framebuffer.nr_samples <= 1) if (sctx->framebuffer.nr_samples <= 1)
si_mark_atom_dirty(sctx, &sctx->msaa_sample_locs.atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_sample_locs);
} }
} }
@ -3409,8 +3409,8 @@ void *si_get_blit_vs(struct si_context *sctx, enum blitter_attrib_type type,
void si_init_shader_functions(struct si_context *sctx) 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->atoms.s.spi_map, si_emit_spi_map);
si_init_atom(sctx, &sctx->scratch_state, &sctx->atoms.s.scratch_state, si_init_atom(sctx, &sctx->atoms.s.scratch_state,
si_emit_scratch_state); si_emit_scratch_state);
sctx->b.create_vs_state = si_create_shader_selector; sctx->b.create_vs_state = si_create_shader_selector;

12
src/gallium/drivers/radeonsi/si_state_streamout.c
View File

@ -83,7 +83,7 @@ void si_streamout_buffers_dirty(struct si_context *sctx)
if (!sctx->streamout.enabled_mask) if (!sctx->streamout.enabled_mask)
return; return;
si_mark_atom_dirty(sctx, &sctx->streamout.begin_atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.streamout_begin);
si_set_streamout_enable(sctx, true); si_set_streamout_enable(sctx, true);
} }
@ -169,7 +169,7 @@ static void si_set_streamout_targets(struct pipe_context *ctx,
if (num_targets) { if (num_targets) {
si_streamout_buffers_dirty(sctx); si_streamout_buffers_dirty(sctx);
} else { } else {
si_set_atom_dirty(sctx, &sctx->streamout.begin_atom, false); si_set_atom_dirty(sctx, &sctx->atoms.s.streamout_begin, false);
si_set_streamout_enable(sctx, false); si_set_streamout_enable(sctx, false);
} }
@ -385,7 +385,7 @@ static void si_set_streamout_enable(struct si_context *sctx, bool enable)
if ((old_strmout_en != si_get_strmout_en(sctx)) || if ((old_strmout_en != si_get_strmout_en(sctx)) ||
(old_hw_enabled_mask != sctx->streamout.hw_enabled_mask)) (old_hw_enabled_mask != sctx->streamout.hw_enabled_mask))
si_mark_atom_dirty(sctx, &sctx->streamout.enable_atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.streamout_enable);
} }
void si_update_prims_generated_query_state(struct si_context *sctx, void si_update_prims_generated_query_state(struct si_context *sctx,
@ -401,7 +401,7 @@ void si_update_prims_generated_query_state(struct si_context *sctx,
sctx->streamout.num_prims_gen_queries != 0; sctx->streamout.num_prims_gen_queries != 0;
if (old_strmout_en != si_get_strmout_en(sctx)) if (old_strmout_en != si_get_strmout_en(sctx))
si_mark_atom_dirty(sctx, &sctx->streamout.enable_atom); si_mark_atom_dirty(sctx, &sctx->atoms.s.streamout_enable);
} }
} }
@ -410,6 +410,6 @@ void si_init_streamout_functions(struct si_context *sctx)
sctx->b.create_stream_output_target = si_create_so_target; sctx->b.create_stream_output_target = si_create_so_target;
sctx->b.stream_output_target_destroy = si_so_target_destroy; sctx->b.stream_output_target_destroy = si_so_target_destroy;
sctx->b.set_stream_output_targets = si_set_streamout_targets; sctx->b.set_stream_output_targets = si_set_streamout_targets;
sctx->streamout.begin_atom.emit = si_emit_streamout_begin; sctx->atoms.s.streamout_begin.emit = si_emit_streamout_begin;
sctx->streamout.enable_atom.emit = si_emit_streamout_enable; sctx->atoms.s.streamout_enable.emit = si_emit_streamout_enable;
} }

18
src/gallium/drivers/radeonsi/si_state_viewport.c
View File

@ -44,7 +44,7 @@ static void si_set_scissor_states(struct pipe_context *pctx,
return; return;
ctx->scissors.dirty_mask |= ((1 << num_scissors) - 1) << start_slot; ctx->scissors.dirty_mask |= ((1 << num_scissors) - 1) << start_slot;
si_mark_atom_dirty(ctx, &ctx->scissors.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
} }
/* Since the guard band disables clipping, we have to clip per-pixel /* Since the guard band disables clipping, we have to clip per-pixel
@ -279,8 +279,8 @@ static void si_set_viewport_states(struct pipe_context *pctx,
ctx->viewports.dirty_mask |= mask; ctx->viewports.dirty_mask |= mask;
ctx->viewports.depth_range_dirty_mask |= mask; ctx->viewports.depth_range_dirty_mask |= mask;
ctx->scissors.dirty_mask |= mask; ctx->scissors.dirty_mask |= mask;
si_mark_atom_dirty(ctx, &ctx->viewports.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
si_mark_atom_dirty(ctx, &ctx->scissors.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
} }
static void si_emit_one_viewport(struct si_context *ctx, static void si_emit_one_viewport(struct si_context *ctx,
@ -415,8 +415,8 @@ void si_update_vs_viewport_state(struct si_context *ctx)
ctx->vs_disables_clipping_viewport = vs_window_space; ctx->vs_disables_clipping_viewport = vs_window_space;
ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1; ctx->viewports.depth_range_dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
si_mark_atom_dirty(ctx, &ctx->scissors.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
si_mark_atom_dirty(ctx, &ctx->viewports.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
} }
/* Viewport index handling. */ /* Viewport index handling. */
@ -425,17 +425,17 @@ void si_update_vs_viewport_state(struct si_context *ctx)
return; return;
if (ctx->scissors.dirty_mask) if (ctx->scissors.dirty_mask)
si_mark_atom_dirty(ctx, &ctx->scissors.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.scissors);
if (ctx->viewports.dirty_mask || if (ctx->viewports.dirty_mask ||
ctx->viewports.depth_range_dirty_mask) ctx->viewports.depth_range_dirty_mask)
si_mark_atom_dirty(ctx, &ctx->viewports.atom); si_mark_atom_dirty(ctx, &ctx->atoms.s.viewports);
} }
void si_init_viewport_functions(struct si_context *ctx) void si_init_viewport_functions(struct si_context *ctx)
{ {
ctx->scissors.atom.emit = si_emit_scissors; ctx->atoms.s.scissors.emit = si_emit_scissors;
ctx->viewports.atom.emit = si_emit_viewport_states; ctx->atoms.s.viewports.emit = si_emit_viewport_states;
ctx->b.set_scissor_states = si_set_scissor_states; ctx->b.set_scissor_states = si_set_scissor_states;
ctx->b.set_viewport_states = si_set_viewport_states; ctx->b.set_viewport_states = si_set_viewport_states;