예제 #1
0
void si_context_gfx_flush(void *context, unsigned flags,
			  struct pipe_fence_handle **fence)
{
	struct si_context *ctx = context;
	struct radeon_winsys_cs *cs = ctx->b.rings.gfx.cs;
	struct radeon_winsys *ws = ctx->b.ws;

	if (cs->cdw == ctx->b.initial_gfx_cs_size &&
	    (!fence || ctx->last_gfx_fence)) {
		if (fence)
			ws->fence_reference(fence, ctx->last_gfx_fence);
		if (!(flags & RADEON_FLUSH_ASYNC))
			ws->cs_sync_flush(cs);
		return;
	}

	ctx->b.rings.gfx.flushing = true;

	r600_preflush_suspend_features(&ctx->b);

	ctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER |
			SI_CONTEXT_INV_TC_L1 |
			SI_CONTEXT_INV_TC_L2 |
			/* this is probably not needed anymore */
			SI_CONTEXT_PS_PARTIAL_FLUSH;
	si_emit_cache_flush(ctx, NULL);

	/* force to keep tiling flags */
	flags |= RADEON_FLUSH_KEEP_TILING_FLAGS;

	if (ctx->trace_buf)
		si_trace_emit(ctx);

	/* Save the IB for debug contexts. */
	if (ctx->is_debug) {
		free(ctx->last_ib);
		ctx->last_ib_dw_size = cs->cdw;
		ctx->last_ib = malloc(cs->cdw * 4);
		memcpy(ctx->last_ib, cs->buf, cs->cdw * 4);
		r600_resource_reference(&ctx->last_trace_buf, ctx->trace_buf);
		r600_resource_reference(&ctx->trace_buf, NULL);
	}

	/* Flush the CS. */
	ws->cs_flush(cs, flags, &ctx->last_gfx_fence,
		     ctx->screen->b.cs_count++);
	ctx->b.rings.gfx.flushing = false;

	if (fence)
		ws->fence_reference(fence, ctx->last_gfx_fence);

	si_begin_new_cs(ctx);
}
예제 #2
0
void si_pm4_emit(struct si_context *sctx, struct si_pm4_state *state)
{
	struct radeon_winsys_cs *cs = sctx->b.rings.gfx.cs;
	for (int i = 0; i < state->nbo; ++i) {
		r600_context_bo_reloc(&sctx->b, &sctx->b.rings.gfx, state->bo[i],
				      state->bo_usage[i], state->bo_priority[i]);
	}

	memcpy(&cs->buf[cs->cdw], state->pm4, state->ndw * 4);

	for (int i = 0; i < state->nrelocs; ++i) {
		cs->buf[cs->cdw + state->relocs[i]] += cs->cdw << 2;
	}

	cs->cdw += state->ndw;

#if SI_TRACE_CS
	if (sctx->screen->b.trace_bo) {
		si_trace_emit(sctx);
	}
#endif
}
예제 #3
0
void si_context_gfx_flush(void *context, unsigned flags,
			  struct pipe_fence_handle **fence)
{
	struct si_context *ctx = context;
	struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
	struct radeon_winsys *ws = ctx->b.ws;

	if (ctx->gfx_flush_in_progress)
		return;

	ctx->gfx_flush_in_progress = true;

	if (cs->cdw == ctx->b.initial_gfx_cs_size &&
	    (!fence || ctx->last_gfx_fence)) {
		if (fence)
			ws->fence_reference(fence, ctx->last_gfx_fence);
		if (!(flags & RADEON_FLUSH_ASYNC))
			ws->cs_sync_flush(cs);
		ctx->gfx_flush_in_progress = false;
		return;
	}

	r600_preflush_suspend_features(&ctx->b);

	ctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER |
			SI_CONTEXT_INV_VMEM_L1 |
			SI_CONTEXT_INV_GLOBAL_L2 |
			/* this is probably not needed anymore */
			SI_CONTEXT_PS_PARTIAL_FLUSH;
	si_emit_cache_flush(ctx, NULL);

	/* force to keep tiling flags */
	flags |= RADEON_FLUSH_KEEP_TILING_FLAGS;

	if (ctx->trace_buf)
		si_trace_emit(ctx);

	if (ctx->is_debug) {
		unsigned i;

		/* Save the IB for debug contexts. */
		free(ctx->last_ib);
		ctx->last_ib_dw_size = cs->cdw;
		ctx->last_ib = malloc(cs->cdw * 4);
		memcpy(ctx->last_ib, cs->buf, cs->cdw * 4);
		r600_resource_reference(&ctx->last_trace_buf, ctx->trace_buf);
		r600_resource_reference(&ctx->trace_buf, NULL);

		/* Save the buffer list. */
		if (ctx->last_bo_list) {
			for (i = 0; i < ctx->last_bo_count; i++)
				pb_reference(&ctx->last_bo_list[i].buf, NULL);
			free(ctx->last_bo_list);
		}
		ctx->last_bo_count = ws->cs_get_buffer_list(cs, NULL);
		ctx->last_bo_list = calloc(ctx->last_bo_count,
					   sizeof(ctx->last_bo_list[0]));
		ws->cs_get_buffer_list(cs, ctx->last_bo_list);
	}

	/* Flush the CS. */
	ws->cs_flush(cs, flags, &ctx->last_gfx_fence,
		     ctx->screen->b.cs_count++);

	if (fence)
		ws->fence_reference(fence, ctx->last_gfx_fence);

	/* Check VM faults if needed. */
	if (ctx->screen->b.debug_flags & DBG_CHECK_VM)
		si_check_vm_faults(ctx);

	si_begin_new_cs(ctx);
	ctx->gfx_flush_in_progress = false;
}
예제 #4
0
void si_begin_new_cs(struct si_context *ctx)
{
	if (ctx->is_debug) {
		uint32_t zero = 0;

		/* Create a buffer used for writing trace IDs and initialize it to 0. */
		assert(!ctx->trace_buf);
		ctx->trace_buf = (struct r600_resource*)
				 pipe_buffer_create(ctx->b.b.screen, PIPE_BIND_CUSTOM,
						    PIPE_USAGE_STAGING, 4);
		if (ctx->trace_buf)
			pipe_buffer_write_nooverlap(&ctx->b.b, &ctx->trace_buf->b.b,
						    0, sizeof(zero), &zero);
		ctx->trace_id = 0;
	}

	if (ctx->trace_buf)
		si_trace_emit(ctx);

	/* Flush read caches at the beginning of CS. */
	ctx->b.flags |= SI_CONTEXT_FLUSH_AND_INV_FRAMEBUFFER |
			SI_CONTEXT_INV_VMEM_L1 |
			SI_CONTEXT_INV_GLOBAL_L2 |
			SI_CONTEXT_INV_SMEM_L1 |
			SI_CONTEXT_INV_ICACHE;

	/* set all valid group as dirty so they get reemited on
	 * next draw command
	 */
	si_pm4_reset_emitted(ctx);

	/* The CS initialization should be emitted before everything else. */
	si_pm4_emit(ctx, ctx->init_config);
	if (ctx->init_config_gs_rings)
		si_pm4_emit(ctx, ctx->init_config_gs_rings);

	ctx->framebuffer.dirty_cbufs = (1 << 8) - 1;
	ctx->framebuffer.dirty_zsbuf = true;
	si_mark_atom_dirty(ctx, &ctx->framebuffer.atom);

	si_mark_atom_dirty(ctx, &ctx->clip_regs);
	si_mark_atom_dirty(ctx, &ctx->clip_state.atom);
	si_mark_atom_dirty(ctx, &ctx->msaa_sample_locs);
	si_mark_atom_dirty(ctx, &ctx->msaa_config);
	si_mark_atom_dirty(ctx, &ctx->sample_mask.atom);
	si_mark_atom_dirty(ctx, &ctx->cb_render_state);
	si_mark_atom_dirty(ctx, &ctx->blend_color.atom);
	si_mark_atom_dirty(ctx, &ctx->db_render_state);
	si_mark_atom_dirty(ctx, &ctx->stencil_ref.atom);
	si_mark_atom_dirty(ctx, &ctx->spi_map);
	si_mark_atom_dirty(ctx, &ctx->b.streamout.enable_atom);
	si_mark_atom_dirty(ctx, &ctx->b.render_cond_atom);
	si_all_descriptors_begin_new_cs(ctx);

	ctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
	ctx->viewports.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
	si_mark_atom_dirty(ctx, &ctx->scissors.atom);
	si_mark_atom_dirty(ctx, &ctx->viewports.atom);

	r600_postflush_resume_features(&ctx->b);

	ctx->b.initial_gfx_cs_size = ctx->b.gfx.cs->cdw;

	/* Invalidate various draw states so that they are emitted before
	 * the first draw call. */
	si_invalidate_draw_sh_constants(ctx);
	ctx->last_primitive_restart_en = -1;
	ctx->last_restart_index = SI_RESTART_INDEX_UNKNOWN;
	ctx->last_gs_out_prim = -1;
	ctx->last_prim = -1;
	ctx->last_multi_vgt_param = -1;
	ctx->last_ls_hs_config = -1;
	ctx->last_rast_prim = -1;
	ctx->last_sc_line_stipple = ~0;
	ctx->emit_scratch_reloc = true;
	ctx->last_ls = NULL;
	ctx->last_tcs = NULL;
	ctx->last_tes_sh_base = -1;
	ctx->last_num_tcs_input_cp = -1;
}