OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

freedreno/a6xx: Split out fd6_state

Browse Source 
Split out the build-up of CP_SET_DRAW_STATE packet, as we are going to
want to re-use this for compute state later when we switch to bindless
IBO descriptors.

While we are at it, drop the enable_mask param, as this is determined
solely by the group_id, and it is easier to maintain a table for the
handful of exceptions to ENABLE_ALL.  The compiler should be able to
optimize away the table lookup.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>
This commit is contained in:
Rob Clark
2023-01-01 08:15:32 -08:00
committed by Marge Bot
parent c73a76195c
commit 1452451185
2 changed files with 103 additions and 91 deletions
Download Patch File Download Diff File Expand all files Collapse all files

113
src/gallium/drivers/freedreno/a6xx/fd6_emit.c
View File

@@ -282,18 +282,15 @@ fd6_emit_combined_textures(struct fd6_emit *emit,
{ {
struct fd_context *ctx = emit->ctx; struct fd_context *ctx = emit->ctx;
static const struct { static const enum fd6_state_id s[PIPE_SHADER_TYPES] = {
enum fd6_state_id state_id; [PIPE_SHADER_VERTEX] = FD6_GROUP_VS_TEX,
unsigned enable_mask; [PIPE_SHADER_TESS_CTRL] = FD6_GROUP_HS_TEX,
} s[PIPE_SHADER_TYPES] = { [PIPE_SHADER_TESS_EVAL] = FD6_GROUP_DS_TEX,
[PIPE_SHADER_VERTEX] = {FD6_GROUP_VS_TEX, ENABLE_ALL}, [PIPE_SHADER_GEOMETRY] = FD6_GROUP_GS_TEX,
[PIPE_SHADER_TESS_CTRL] = {FD6_GROUP_HS_TEX, ENABLE_ALL}, [PIPE_SHADER_FRAGMENT] = FD6_GROUP_FS_TEX,
[PIPE_SHADER_TESS_EVAL] = {FD6_GROUP_DS_TEX, ENABLE_ALL},
[PIPE_SHADER_GEOMETRY] = {FD6_GROUP_GS_TEX, ENABLE_ALL},
[PIPE_SHADER_FRAGMENT] = {FD6_GROUP_FS_TEX, ENABLE_DRAW},
}; };
assert(s[type].state_id); assert((type < ARRAY_SIZE(s)) && s[type]);
if (!v->image_mapping.num_tex && !v->fb_read) { if (!v->image_mapping.num_tex && !v->fb_read) {
/* in the fast-path, when we don't have to mix in any image/SSBO /* in the fast-path, when we don't have to mix in any image/SSBO
@@ -303,13 +300,11 @@ fd6_emit_combined_textures(struct fd6_emit *emit,
* Also, framebuffer-read is a slow-path because an extra * Also, framebuffer-read is a slow-path because an extra
* texture needs to be inserted. * texture needs to be inserted.
*/ */
if ((ctx->dirty_shader[type] & FD_DIRTY_SHADER_TEX) && if (ctx->tex[type].num_textures > 0) {
ctx->tex[type].num_textures > 0) {
struct fd6_texture_state *tex = struct fd6_texture_state *tex =
fd6_texture_state(ctx, type, &ctx->tex[type]); fd6_texture_state(ctx, type, &ctx->tex[type]);
fd6_emit_add_group(emit, tex->stateobj, s[type].state_id, fd6_state_add_group(&emit->state, tex->stateobj, s[type]);
s[type].enable_mask);
fd6_texture_state_reference(&tex, NULL); fd6_texture_state_reference(&tex, NULL);
} }
@@ -327,8 +322,7 @@ fd6_emit_combined_textures(struct fd6_emit *emit,
fd6_emit_textures(ctx, stateobj, type, tex, v); fd6_emit_textures(ctx, stateobj, type, tex, v);
fd6_emit_take_group(emit, stateobj, s[type].state_id, fd6_state_take_group(&emit->state, stateobj, s[type]);
s[type].enable_mask);
} }
} }
} }
@@ -705,14 +699,13 @@ fd6_emit_streamout(struct fd_ringbuffer *ring, struct fd6_emit *emit) assert_dt
} }
if (emit->streamout_mask) { if (emit->streamout_mask) {
fd6_emit_add_group(emit, prog->streamout_stateobj, FD6_GROUP_SO, fd6_state_add_group(&emit->state, prog->streamout_stateobj, FD6_GROUP_SO);
ENABLE_ALL);
} else if (ctx->last.streamout_mask != 0) { } else if (ctx->last.streamout_mask != 0) {
/* If we transition from a draw with streamout to one without, turn /* If we transition from a draw with streamout to one without, turn
* off streamout. * off streamout.
*/ */
fd6_emit_add_group(emit, fd6_context(ctx)->streamout_disable_stateobj, fd6_state_add_group(&emit->state, fd6_context(ctx)->streamout_disable_stateobj,
FD6_GROUP_SO, ENABLE_ALL); FD6_GROUP_SO);
} }
/* Make sure that any use of our TFB outputs (indirect draw source or shader /* Make sure that any use of our TFB outputs (indirect draw source or shader
@@ -824,131 +817,111 @@ fd6_emit_3d_state(struct fd_ringbuffer *ring, struct fd6_emit *emit)
u_foreach_bit (b, emit->dirty_groups) { u_foreach_bit (b, emit->dirty_groups) {
enum fd6_state_id group = b; enum fd6_state_id group = b;
struct fd_ringbuffer *state = NULL; struct fd_ringbuffer *state = NULL;
uint32_t enable_mask = ENABLE_ALL;
switch (group) { switch (group) {
case FD6_GROUP_VTXSTATE: case FD6_GROUP_VTXSTATE:
state = fd6_vertex_stateobj(ctx->vtx.vtx)->stateobj; state = fd6_vertex_stateobj(ctx->vtx.vtx)->stateobj;
fd_ringbuffer_ref(state); fd6_state_add_group(&emit->state, state, FD6_GROUP_VTXSTATE);
break; break;
case FD6_GROUP_VBO: case FD6_GROUP_VBO:
state = build_vbo_state(emit); state = build_vbo_state(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_VBO);
break; break;
case FD6_GROUP_ZSA: case FD6_GROUP_ZSA:
state = fd6_zsa_state( state = fd6_zsa_state(
ctx, ctx,
util_format_is_pure_integer(pipe_surface_format(pfb->cbufs[0])), util_format_is_pure_integer(pipe_surface_format(pfb->cbufs[0])),
fd_depth_clamp_enabled(ctx)); fd_depth_clamp_enabled(ctx));
fd_ringbuffer_ref(state); fd6_state_add_group(&emit->state, state, FD6_GROUP_ZSA);
break; break;
case FD6_GROUP_LRZ: case FD6_GROUP_LRZ:
state = build_lrz(emit); state = build_lrz(emit);
if (!state) if (state)
continue; fd6_state_take_group(&emit->state, state, FD6_GROUP_LRZ);
break; break;
case FD6_GROUP_SCISSOR: case FD6_GROUP_SCISSOR:
state = build_scissor(emit); state = build_scissor(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_SCISSOR);
break; break;
case FD6_GROUP_PROG: case FD6_GROUP_PROG:
fd6_emit_add_group(emit, prog->config_stateobj, FD6_GROUP_PROG_CONFIG, fd6_state_add_group(&emit->state, prog->config_stateobj,
ENABLE_ALL); FD6_GROUP_PROG_CONFIG);
fd6_emit_add_group(emit, prog->stateobj, FD6_GROUP_PROG, ENABLE_DRAW); fd6_state_add_group(&emit->state, prog->stateobj, FD6_GROUP_PROG);
fd6_emit_add_group(emit, prog->binning_stateobj, fd6_state_add_group(&emit->state, prog->binning_stateobj,
FD6_GROUP_PROG_BINNING, FD6_GROUP_PROG_BINNING);
CP_SET_DRAW_STATE__0_BINNING);
/* emit remaining streaming program state, ie. what depends on /* emit remaining streaming program state, ie. what depends on
* other emit state, so cannot be pre-baked. * other emit state, so cannot be pre-baked.
*/ */
fd6_emit_take_group(emit, fd6_program_interp_state(emit), fd6_state_take_group(&emit->state, fd6_program_interp_state(emit),
FD6_GROUP_PROG_INTERP, ENABLE_DRAW); FD6_GROUP_PROG_INTERP);
continue; break;
case FD6_GROUP_RASTERIZER: case FD6_GROUP_RASTERIZER:
state = fd6_rasterizer_state(ctx, emit->primitive_restart); state = fd6_rasterizer_state(ctx, emit->primitive_restart);
fd_ringbuffer_ref(state); fd6_state_add_group(&emit->state, state, FD6_GROUP_RASTERIZER);
break; break;
case FD6_GROUP_PROG_FB_RAST: case FD6_GROUP_PROG_FB_RAST:
state = build_prog_fb_rast(emit); state = build_prog_fb_rast(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_PROG_FB_RAST);
break; break;
case FD6_GROUP_BLEND: case FD6_GROUP_BLEND:
state = fd6_blend_variant(ctx->blend, pfb->samples, ctx->sample_mask) state = fd6_blend_variant(ctx->blend, pfb->samples, ctx->sample_mask)
->stateobj; ->stateobj;
fd_ringbuffer_ref(state); fd6_state_add_group(&emit->state, state, FD6_GROUP_BLEND);
break; break;
case FD6_GROUP_BLEND_COLOR: case FD6_GROUP_BLEND_COLOR:
state = build_blend_color(emit); state = build_blend_color(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_BLEND_COLOR);
break; break;
case FD6_GROUP_IBO: case FD6_GROUP_IBO:
state = build_ibo(emit); state = build_ibo(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_IBO);
break; break;
case FD6_GROUP_CONST: case FD6_GROUP_CONST:
state = fd6_build_user_consts(emit); state = fd6_build_user_consts(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_CONST);
break; break;
case FD6_GROUP_DRIVER_PARAMS: case FD6_GROUP_DRIVER_PARAMS:
state = fd6_build_driver_params(emit); state = fd6_build_driver_params(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_DRIVER_PARAMS);
break; break;
case FD6_GROUP_PRIMITIVE_PARAMS: case FD6_GROUP_PRIMITIVE_PARAMS:
state = fd6_build_tess_consts(emit); state = fd6_build_tess_consts(emit);
fd6_state_take_group(&emit->state, state, FD6_GROUP_PRIMITIVE_PARAMS);
break; break;
case FD6_GROUP_VS_TEX: case FD6_GROUP_VS_TEX:
fd6_emit_combined_textures(emit, PIPE_SHADER_VERTEX, vs); fd6_emit_combined_textures(emit, PIPE_SHADER_VERTEX, vs);
continue; break;
case FD6_GROUP_HS_TEX: case FD6_GROUP_HS_TEX:
if (hs) { if (hs) {
fd6_emit_combined_textures(emit, PIPE_SHADER_TESS_CTRL, hs); fd6_emit_combined_textures(emit, PIPE_SHADER_TESS_CTRL, hs);
} }
continue; break;
case FD6_GROUP_DS_TEX: case FD6_GROUP_DS_TEX:
if (ds) { if (ds) {
fd6_emit_combined_textures(emit, PIPE_SHADER_TESS_EVAL, ds); fd6_emit_combined_textures(emit, PIPE_SHADER_TESS_EVAL, ds);
} }
continue; break;
case FD6_GROUP_GS_TEX: case FD6_GROUP_GS_TEX:
if (gs) { if (gs) {
fd6_emit_combined_textures(emit, PIPE_SHADER_GEOMETRY, gs); fd6_emit_combined_textures(emit, PIPE_SHADER_GEOMETRY, gs);
} }
continue; break;
case FD6_GROUP_FS_TEX: case FD6_GROUP_FS_TEX:
fd6_emit_combined_textures(emit, PIPE_SHADER_FRAGMENT, fs); fd6_emit_combined_textures(emit, PIPE_SHADER_FRAGMENT, fs);
continue; break;
case FD6_GROUP_SO: case FD6_GROUP_SO:
fd6_emit_streamout(ring, emit); fd6_emit_streamout(ring, emit);
continue; break;
case FD6_GROUP_NON_GROUP: case FD6_GROUP_NON_GROUP:
fd6_emit_non_ring(ring, emit); fd6_emit_non_ring(ring, emit);
continue; break;
default: default:
unreachable("bad state group"); unreachable("bad state group");
} }
fd6_emit_take_group(emit, state, group, enable_mask);
} }
if (emit->num_groups > 0) { fd6_state_emit(&emit->state, ring);
OUT_PKT7(ring, CP_SET_DRAW_STATE, 3 * emit->num_groups);
for (unsigned i = 0; i < emit->num_groups; i++) {
struct fd6_state_group *g = &emit->groups[i];
unsigned n = g->stateobj ? fd_ringbuffer_size(g->stateobj) / 4 : 0;
assert((g->enable_mask & ~ENABLE_ALL) == 0);
if (n == 0) {
OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(0) |
CP_SET_DRAW_STATE__0_DISABLE | g->enable_mask |
CP_SET_DRAW_STATE__0_GROUP_ID(g->group_id));
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
} else {
OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(n) | g->enable_mask |
CP_SET_DRAW_STATE__0_GROUP_ID(g->group_id));
OUT_RB(ring, g->stateobj);
}
if (g->stateobj)
fd_ringbuffer_del(g->stateobj);
}
emit->num_groups = 0;
}
} }
void void

81
src/gallium/drivers/freedreno/a6xx/fd6_emit.h
View File

@@ -83,6 +83,65 @@ struct fd6_state_group {
uint32_t enable_mask; uint32_t enable_mask;
}; };
struct fd6_state {
struct fd6_state_group groups[32];
unsigned num_groups;
};
static inline void
fd6_state_emit(struct fd6_state *state, struct fd_ringbuffer *ring)
{
if (!state->num_groups)
return;
OUT_PKT7(ring, CP_SET_DRAW_STATE, 3 * state->num_groups);
for (unsigned i = 0; i < state->num_groups; i++) {
struct fd6_state_group *g = &state->groups[i];
unsigned n = g->stateobj ? fd_ringbuffer_size(g->stateobj) / 4 : 0;
assert((g->enable_mask & ~ENABLE_ALL) == 0);
if (n == 0) {
OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(0) |
CP_SET_DRAW_STATE__0_DISABLE | g->enable_mask |
CP_SET_DRAW_STATE__0_GROUP_ID(g->group_id));
OUT_RING(ring, 0x00000000);
OUT_RING(ring, 0x00000000);
} else {
OUT_RING(ring, CP_SET_DRAW_STATE__0_COUNT(n) | g->enable_mask |
CP_SET_DRAW_STATE__0_GROUP_ID(g->group_id));
OUT_RB(ring, g->stateobj);
}
if (g->stateobj)
fd_ringbuffer_del(g->stateobj);
}
}
static inline void
fd6_state_take_group(struct fd6_state *state, struct fd_ringbuffer *stateobj,
enum fd6_state_id group_id)
{
static const unsigned enable_mask[32] = {
[FD6_GROUP_PROG] = ENABLE_DRAW,
[FD6_GROUP_PROG_BINNING] = CP_SET_DRAW_STATE__0_BINNING,
[FD6_GROUP_PROG_INTERP] = ENABLE_DRAW,
[FD6_GROUP_FS_TEX] = ENABLE_DRAW,
};
assert(state->num_groups < ARRAY_SIZE(state->groups));
struct fd6_state_group *g = &state->groups[state->num_groups++];
g->stateobj = stateobj;
g->group_id = group_id;
g->enable_mask = enable_mask[group_id] ? enable_mask[group_id] : ENABLE_ALL;
}
static inline void
fd6_state_add_group(struct fd6_state *state, struct fd_ringbuffer *stateobj,
enum fd6_state_id group_id)
{
fd6_state_take_group(state, fd_ringbuffer_ref(stateobj), group_id);
}
/* grouped together emit-state for prog/vertex/state emit: */ /* grouped together emit-state for prog/vertex/state emit: */
struct fd6_emit { struct fd6_emit {
struct fd_context *ctx; struct fd_context *ctx;
@@ -112,8 +171,7 @@ struct fd6_emit {
unsigned streamout_mask; unsigned streamout_mask;
struct fd6_state_group groups[32]; struct fd6_state state;
unsigned num_groups;
}; };
static inline const struct fd6_program_state * static inline const struct fd6_program_state *
@@ -127,25 +185,6 @@ fd6_emit_get_prog(struct fd6_emit *emit)
return emit->prog; return emit->prog;
} }
static inline void
fd6_emit_take_group(struct fd6_emit *emit, struct fd_ringbuffer *stateobj,
enum fd6_state_id group_id, unsigned enable_mask)
{
assert(emit->num_groups < ARRAY_SIZE(emit->groups));
struct fd6_state_group *g = &emit->groups[emit->num_groups++];
g->stateobj = stateobj;
g->group_id = group_id;
g->enable_mask = enable_mask;
}
static inline void
fd6_emit_add_group(struct fd6_emit *emit, struct fd_ringbuffer *stateobj,
enum fd6_state_id group_id, unsigned enable_mask)
{
fd6_emit_take_group(emit, fd_ringbuffer_ref(stateobj), group_id,
enable_mask);
}
static inline unsigned static inline unsigned
fd6_event_write(struct fd_batch *batch, struct fd_ringbuffer *ring, fd6_event_write(struct fd_batch *batch, struct fd_ringbuffer *ring,
enum vgt_event_type evt, bool timestamp) enum vgt_event_type evt, bool timestamp)