blend &&
blend->blend_enable_4bit & cb_target_mask &&
sctx->framebuffer.nr_samples >= 2;
+ unsigned watermark = sctx->framebuffer.dcc_overwrite_combiner_watermark;
radeon_opt_set_context_reg(
sctx, R_028424_CB_DCC_CONTROL,
SI_TRACKED_CB_DCC_CONTROL,
S_028424_OVERWRITE_COMBINER_MRT_SHARING_DISABLE(1) |
- S_028424_OVERWRITE_COMBINER_WATERMARK(4) |
- S_028424_OVERWRITE_COMBINER_DISABLE(oc_disable));
+ S_028424_OVERWRITE_COMBINER_WATERMARK(watermark) |
+ S_028424_OVERWRITE_COMBINER_DISABLE(oc_disable) |
+ S_028424_DISABLE_CONSTANT_ENCODE_REG(sctx->screen->has_dcc_constant_encode));
}
/* RB+ register settings. */
si_pm4_set_reg(pm4, R_028B70_DB_ALPHA_TO_MASK,
S_028B70_ALPHA_TO_MASK_ENABLE(state->alpha_to_coverage) |
- S_028B70_ALPHA_TO_MASK_OFFSET0(2) |
- S_028B70_ALPHA_TO_MASK_OFFSET1(2) |
- S_028B70_ALPHA_TO_MASK_OFFSET2(2) |
- S_028B70_ALPHA_TO_MASK_OFFSET3(2));
+ S_028B70_ALPHA_TO_MASK_OFFSET0(3) |
+ S_028B70_ALPHA_TO_MASK_OFFSET1(1) |
+ S_028B70_ALPHA_TO_MASK_OFFSET2(0) |
+ S_028B70_ALPHA_TO_MASK_OFFSET3(2) |
+ S_028B70_OFFSET_ROUND(1));
if (state->alpha_to_coverage)
blend->need_src_alpha_4bit |= 0xf;
if (state->point_size_per_vertex) {
psize_min = util_get_min_point_size(state);
- psize_max = 8192;
+ psize_max = SI_MAX_POINT_SIZE;
} else {
/* Force the point size to be as if the vertex output was disabled. */
psize_min = state->point_size;
si_get_pipe_constant_buffer(sctx, PIPE_SHADER_COMPUTE, 0, &st->saved_const0);
si_get_shader_buffers(sctx, PIPE_SHADER_COMPUTE, 0, 3, st->saved_ssbo);
+
+ st->saved_ssbo_writable_mask = 0;
+
+ for (unsigned i = 0; i < 3; i++) {
+ if (sctx->const_and_shader_buffers[PIPE_SHADER_COMPUTE].writable_mask &
+ (1u << si_get_shaderbuf_slot(i)))
+ st->saved_ssbo_writable_mask |= 1 << i;
+ }
}
static void si_emit_db_render_state(struct si_context *sctx)
unsigned retval = 0;
if (target >= PIPE_MAX_TEXTURE_TYPES) {
- PRINT_ERR("r600: unsupported texture type %d\n", target);
+ PRINT_ERR("radeonsi: unsupported texture type %d\n", target);
return false;
}
*
* Only flush and wait for CB if there is actually a bound color buffer.
*/
- if (sctx->framebuffer.uncompressed_cb_mask)
+ if (sctx->framebuffer.uncompressed_cb_mask) {
si_make_CB_shader_coherent(sctx, sctx->framebuffer.nr_samples,
- sctx->framebuffer.CB_has_shader_readable_metadata);
+ sctx->framebuffer.CB_has_shader_readable_metadata,
+ sctx->framebuffer.all_DCC_pipe_aligned);
+ }
sctx->flags |= SI_CONTEXT_CS_PARTIAL_FLUSH;
sctx->framebuffer.any_dst_linear = false;
sctx->framebuffer.CB_has_shader_readable_metadata = false;
sctx->framebuffer.DB_has_shader_readable_metadata = false;
+ sctx->framebuffer.all_DCC_pipe_aligned = true;
+ unsigned num_bpp64_colorbufs = 0;
for (i = 0; i < state->nr_cbufs; i++) {
if (!state->cbufs[i])
if (tex->surface.is_linear)
sctx->framebuffer.any_dst_linear = true;
+ if (tex->surface.bpe >= 8)
+ num_bpp64_colorbufs++;
- if (vi_dcc_enabled(tex, surf->base.u.tex.level))
+ if (vi_dcc_enabled(tex, surf->base.u.tex.level)) {
sctx->framebuffer.CB_has_shader_readable_metadata = true;
+ if (sctx->chip_class >= GFX9 &&
+ !tex->surface.u.gfx9.dcc.pipe_aligned)
+ sctx->framebuffer.all_DCC_pipe_aligned = false;
+ }
+
si_context_add_resource_size(sctx, surf->base.texture);
p_atomic_inc(&tex->framebuffers_bound);
}
}
+ /* For optimal DCC performance. */
+ if (sctx->chip_class == VI)
+ sctx->framebuffer.dcc_overwrite_combiner_watermark = 4;
+ else if (num_bpp64_colorbufs >= 5)
+ sctx->framebuffer.dcc_overwrite_combiner_watermark = 8;
+ else
+ sctx->framebuffer.dcc_overwrite_combiner_watermark = 6;
+
struct si_texture *zstex = NULL;
if (state->zsbuf) {
* @param state 256-bit descriptor; only the high 128 bits are filled in
*/
void
-si_make_buffer_descriptor(struct si_screen *screen, struct r600_resource *buf,
+si_make_buffer_descriptor(struct si_screen *screen, struct si_resource *buf,
enum pipe_format format,
unsigned offset, unsigned size,
uint32_t *state)
* - For VMEM and inst.IDXEN == 0 or STRIDE == 0, it's in byte units.
* - For VMEM and inst.IDXEN == 1 and STRIDE != 0, it's in units of STRIDE.
*/
- if (screen->info.chip_class >= GFX9)
- /* When vindex == 0, LLVM sets IDXEN = 0, thus changing units
+ if (screen->info.chip_class >= GFX9 && HAVE_LLVM < 0x0800)
+ /* When vindex == 0, LLVM < 8.0 sets IDXEN = 0, thus changing units
* from STRIDE to bytes. This works around it by setting
* NUM_RECORDS to at least the size of one element, so that
* the first element is readable when IDXEN == 0.
- *
- * TODO: Fix this in LLVM, but do we need a new intrinsic where
- * IDXEN is enforced?
*/
num_records = num_records ? MAX2(num_records, stride) : 0;
else if (screen->info.chip_class == VI)
/* Buffer resource. */
if (texture->target == PIPE_BUFFER) {
si_make_buffer_descriptor(sctx->screen,
- r600_resource(texture),
+ si_resource(texture),
state->format,
state->u.buf.offset,
state->u.buf.size,
unsigned num_divisors = util_last_bit(v->instance_divisor_is_fetched);
v->instance_divisor_factor_buffer =
- (struct r600_resource*)
+ (struct si_resource*)
pipe_buffer_create(&sscreen->b, 0, PIPE_USAGE_DEFAULT,
num_divisors * sizeof(divisor_factors[0]));
if (!v->instance_divisor_factor_buffer) {
if (sctx->vertex_elements == state)
sctx->vertex_elements = NULL;
- r600_resource_reference(&v->instance_divisor_factor_buffer, NULL);
+ si_resource_reference(&v->instance_divisor_factor_buffer, NULL);
FREE(state);
}
dsti->stride = src->stride;
si_context_add_resource_size(sctx, buf);
if (buf)
- r600_resource(buf)->bind_history |= PIPE_BIND_VERTEX_BUFFER;
+ si_resource(buf)->bind_history |= PIPE_BIND_VERTEX_BUFFER;
}
} else {
for (i = 0; i < count; i++) {
cb.user_buffer = NULL;
cb.buffer_size = sizeof(array);
- si_upload_const_buffer(sctx, (struct r600_resource**)&cb.buffer,
+ si_upload_const_buffer(sctx, (struct si_resource**)&cb.buffer,
(void*)array, sizeof(array),
&cb.buffer_offset);
si_update_fb_dirtiness_after_rendering(sctx);
/* Multisample surfaces are flushed in si_decompress_textures. */
- if (sctx->framebuffer.uncompressed_cb_mask)
+ if (sctx->framebuffer.uncompressed_cb_mask) {
si_make_CB_shader_coherent(sctx, sctx->framebuffer.nr_samples,
- sctx->framebuffer.CB_has_shader_readable_metadata);
+ sctx->framebuffer.CB_has_shader_readable_metadata,
+ sctx->framebuffer.all_DCC_pipe_aligned);
+ }
}
/* This only ensures coherency for shader image/buffer stores. */
-static void si_memory_barrier(struct pipe_context *ctx, unsigned flags)
+void si_memory_barrier(struct pipe_context *ctx, unsigned flags)
{
struct si_context *sctx = (struct si_context *)ctx;
+ if (!(flags & ~PIPE_BARRIER_UPDATE))
+ return;
+
/* Subsequent commands must wait for all shader invocations to
* complete. */
sctx->flags |= SI_CONTEXT_PS_PARTIAL_FLUSH |
sctx->b.set_vertex_buffers = si_set_vertex_buffers;
sctx->b.texture_barrier = si_texture_barrier;
- sctx->b.memory_barrier = si_memory_barrier;
sctx->b.set_min_samples = si_set_min_samples;
sctx->b.set_tess_state = si_set_tess_state;
sctx->b.set_active_query_state = si_set_active_query_state;
- sctx->b.draw_vbo = si_draw_vbo;
-
si_init_config(sctx);
}
bool has_clear_state = sscreen->has_clear_state;
struct si_pm4_state *pm4 = CALLOC_STRUCT(si_pm4_state);
- /* Only SI can disable CLEAR_STATE for now. */
- assert(has_clear_state || sscreen->info.chip_class == SI);
+ /* SI, radeon kernel disabled CLEAR_STATE. */
+ assert(has_clear_state || sscreen->info.chip_class == SI ||
+ sscreen->info.drm_major != 3);
if (!pm4)
return;
pc_lines = 4096;
break;
case CHIP_RAVEN:
+ case CHIP_RAVEN2:
pc_lines = 1024;
break;
default:
projects / mesa.git / blobdiff