[Mesa-dev] [PATCH 13/15] radeonsi: don't re-upload the sample position constant buffer repeatedly
Marek Olšák
maraeo at gmail.com
Tue Oct 2 22:35:45 UTC 2018
From: Marek Olšák <marek.olsak at amd.com>
---
src/gallium/drivers/radeonsi/si_pipe.c | 7 +++++++
src/gallium/drivers/radeonsi/si_pipe.h | 13 ++++++++-----
src/gallium/drivers/radeonsi/si_state.c | 19 +++++++++++++------
src/gallium/drivers/radeonsi/si_state_msaa.c | 10 +++++-----
4 files changed, 33 insertions(+), 16 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_pipe.c b/src/gallium/drivers/radeonsi/si_pipe.c
index 4da361c42ee..5ae9c298e77 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.c
+++ b/src/gallium/drivers/radeonsi/si_pipe.c
@@ -153,20 +153,21 @@ static void si_destroy_context(struct pipe_context *context)
struct pipe_framebuffer_state fb = {};
if (context->set_framebuffer_state)
context->set_framebuffer_state(context, &fb);
si_release_all_descriptors(sctx);
pipe_resource_reference(&sctx->esgs_ring, NULL);
pipe_resource_reference(&sctx->gsvs_ring, NULL);
pipe_resource_reference(&sctx->tess_rings, NULL);
pipe_resource_reference(&sctx->null_const_buf.buffer, NULL);
+ pipe_resource_reference(&sctx->sample_pos_buffer, NULL);
r600_resource_reference(&sctx->border_color_buffer, NULL);
free(sctx->border_color_table);
r600_resource_reference(&sctx->scratch_buffer, NULL);
r600_resource_reference(&sctx->compute_scratch_buffer, NULL);
r600_resource_reference(&sctx->wait_mem_scratch, NULL);
si_pm4_free_state(sctx, sctx->init_config, ~0);
if (sctx->init_config_gs_rings)
si_pm4_free_state(sctx, sctx->init_config_gs_rings, ~0);
for (i = 0; i < ARRAY_SIZE(sctx->vgt_shader_config); i++)
@@ -592,20 +593,26 @@ static struct pipe_context *si_create_context(struct pipe_screen *screen,
_mesa_key_pointer_equal);
sctx->img_handles = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
util_dynarray_init(&sctx->resident_tex_handles, NULL);
util_dynarray_init(&sctx->resident_img_handles, NULL);
util_dynarray_init(&sctx->resident_tex_needs_color_decompress, NULL);
util_dynarray_init(&sctx->resident_img_needs_color_decompress, NULL);
util_dynarray_init(&sctx->resident_tex_needs_depth_decompress, NULL);
+ sctx->sample_pos_buffer =
+ pipe_buffer_create(sctx->b.screen, 0, PIPE_USAGE_DEFAULT,
+ sizeof(sctx->sample_positions));
+ pipe_buffer_write(&sctx->b, sctx->sample_pos_buffer, 0,
+ sizeof(sctx->sample_positions), &sctx->sample_positions);
+
/* this must be last */
si_begin_new_gfx_cs(sctx);
return &sctx->b;
fail:
fprintf(stderr, "radeonsi: Failed to create a context.\n");
si_destroy_context(&sctx->b);
return NULL;
}
static struct pipe_context *si_pipe_create_context(struct pipe_screen *screen,
diff --git a/src/gallium/drivers/radeonsi/si_pipe.h b/src/gallium/drivers/radeonsi/si_pipe.h
index ff11eab0224..93082e262d6 100644
--- a/src/gallium/drivers/radeonsi/si_pipe.h
+++ b/src/gallium/drivers/radeonsi/si_pipe.h
@@ -958,25 +958,28 @@ struct si_context {
struct util_dynarray resident_img_needs_color_decompress;
struct util_dynarray resident_tex_needs_depth_decompress;
/* Bindless state */
bool uses_bindless_samplers;
bool uses_bindless_images;
/* MSAA sample locations.
* The first index is the sample index.
* The second index is the coordinate: X, Y. */
- float sample_locations_1x[1][2];
- float sample_locations_2x[2][2];
- float sample_locations_4x[4][2];
- float sample_locations_8x[8][2];
- float sample_locations_16x[16][2];
+ struct {
+ float x1[1][2];
+ float x2[2][2];
+ float x4[4][2];
+ float x8[8][2];
+ float x16[16][2];
+ } sample_positions;
+ struct pipe_resource *sample_pos_buffer;
/* Misc stats. */
unsigned num_draw_calls;
unsigned num_decompress_calls;
unsigned num_mrt_draw_calls;
unsigned num_prim_restart_calls;
unsigned num_spill_draw_calls;
unsigned num_compute_calls;
unsigned num_spill_compute_calls;
unsigned num_dma_calls;
diff --git a/src/gallium/drivers/radeonsi/si_state.c b/src/gallium/drivers/radeonsi/si_state.c
index 0cebd974d80..f4fc4fd69da 100644
--- a/src/gallium/drivers/radeonsi/si_state.c
+++ b/src/gallium/drivers/radeonsi/si_state.c
@@ -2711,21 +2711,20 @@ static void si_dec_framebuffer_counters(const struct pipe_framebuffer_state *sta
tex = (struct si_texture*)surf->base.texture;
p_atomic_dec(&tex->framebuffers_bound);
}
}
static void si_set_framebuffer_state(struct pipe_context *ctx,
const struct pipe_framebuffer_state *state)
{
struct si_context *sctx = (struct si_context *)ctx;
- struct pipe_constant_buffer constbuf = {0};
struct si_surface *surf = NULL;
struct si_texture *tex;
bool old_any_dst_linear = sctx->framebuffer.any_dst_linear;
unsigned old_nr_samples = sctx->framebuffer.nr_samples;
unsigned old_colorbuf_enabled_4bit = sctx->framebuffer.colorbuf_enabled_4bit;
bool old_has_zsbuf = !!sctx->framebuffer.state.zsbuf;
bool old_has_stencil =
old_has_zsbuf &&
((struct si_texture*)sctx->framebuffer.state.zsbuf->texture)->surface.has_stencil;
bool unbound = false;
@@ -2935,39 +2934,47 @@ static void si_set_framebuffer_state(struct pipe_context *ctx,
if (sctx->framebuffer.any_dst_linear != old_any_dst_linear)
si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
if (sctx->screen->has_out_of_order_rast &&
(sctx->framebuffer.colorbuf_enabled_4bit != old_colorbuf_enabled_4bit ||
!!sctx->framebuffer.state.zsbuf != old_has_zsbuf ||
(zstex && zstex->surface.has_stencil != old_has_stencil)))
si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
if (sctx->framebuffer.nr_samples != old_nr_samples) {
+ struct pipe_constant_buffer constbuf = {0};
+
si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_config);
si_mark_atom_dirty(sctx, &sctx->atoms.s.db_render_state);
+ constbuf.buffer = sctx->sample_pos_buffer;
+
/* Set sample locations as fragment shader constants. */
switch (sctx->framebuffer.nr_samples) {
case 1:
- constbuf.user_buffer = sctx->sample_locations_1x;
+ constbuf.buffer_offset = 0;
break;
case 2:
- constbuf.user_buffer = sctx->sample_locations_2x;
+ constbuf.buffer_offset = (ubyte*)sctx->sample_positions.x2 -
+ (ubyte*)sctx->sample_positions.x1;
break;
case 4:
- constbuf.user_buffer = sctx->sample_locations_4x;
+ constbuf.buffer_offset = (ubyte*)sctx->sample_positions.x4 -
+ (ubyte*)sctx->sample_positions.x1;
break;
case 8:
- constbuf.user_buffer = sctx->sample_locations_8x;
+ constbuf.buffer_offset = (ubyte*)sctx->sample_positions.x8 -
+ (ubyte*)sctx->sample_positions.x1;
break;
case 16:
- constbuf.user_buffer = sctx->sample_locations_16x;
+ constbuf.buffer_offset = (ubyte*)sctx->sample_positions.x16 -
+ (ubyte*)sctx->sample_positions.x1;
break;
default:
PRINT_ERR("Requested an invalid number of samples %i.\n",
sctx->framebuffer.nr_samples);
assert(0);
}
constbuf.buffer_size = sctx->framebuffer.nr_samples * 2 * 4;
si_set_rw_buffer(sctx, SI_PS_CONST_SAMPLE_POSITIONS, &constbuf);
si_mark_atom_dirty(sctx, &sctx->atoms.s.msaa_sample_locs);
diff --git a/src/gallium/drivers/radeonsi/si_state_msaa.c b/src/gallium/drivers/radeonsi/si_state_msaa.c
index f9387e75ed1..b741bcadec8 100644
--- a/src/gallium/drivers/radeonsi/si_state_msaa.c
+++ b/src/gallium/drivers/radeonsi/si_state_msaa.c
@@ -191,21 +191,21 @@ void si_emit_sample_locations(struct radeon_cmdbuf *cs, int nr_samples)
break;
}
}
void si_init_msaa_functions(struct si_context *sctx)
{
int i;
sctx->b.get_sample_position = si_get_sample_position;
- si_get_sample_position(&sctx->b, 1, 0, sctx->sample_locations_1x[0]);
+ si_get_sample_position(&sctx->b, 1, 0, sctx->sample_positions.x1[0]);
for (i = 0; i < 2; i++)
- si_get_sample_position(&sctx->b, 2, i, sctx->sample_locations_2x[i]);
+ si_get_sample_position(&sctx->b, 2, i, sctx->sample_positions.x2[i]);
for (i = 0; i < 4; i++)
- si_get_sample_position(&sctx->b, 4, i, sctx->sample_locations_4x[i]);
+ si_get_sample_position(&sctx->b, 4, i, sctx->sample_positions.x4[i]);
for (i = 0; i < 8; i++)
- si_get_sample_position(&sctx->b, 8, i, sctx->sample_locations_8x[i]);
+ si_get_sample_position(&sctx->b, 8, i, sctx->sample_positions.x8[i]);
for (i = 0; i < 16; i++)
- si_get_sample_position(&sctx->b, 16, i, sctx->sample_locations_16x[i]);
+ si_get_sample_position(&sctx->b, 16, i, sctx->sample_positions.x16[i]);
}
--
2.17.1
More information about the mesa-dev
mailing list