summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/radeonsi/si_hw_context.c
blob: 92d0cc51dd66049b9a1d592bda5b95140c315b51 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
/*
 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * on the rights to use, copy, modify, merge, publish, distribute, sub
 * license, and/or sell copies of the Software, and to permit persons to whom
 * the Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
 * USE OR OTHER DEALINGS IN THE SOFTWARE.
 *
 * Authors:
 *      Jerome Glisse
 */

#include "si_pipe.h"
#include "radeon/r600_cs.h"

static unsigned si_descriptor_list_cs_space(unsigned count, unsigned element_size)
{
	/* Ensure we have enough space to start a new range in a hole */
	assert(element_size >= 3);

	/* 5 dwords for write to L2 + 3 bytes for the packet header of
	 * every disjoint range written to CE RAM.
	 */
	return 5 + (3 * count / 2) + count * element_size;
}

static unsigned si_ce_needed_cs_space(void)
{
	unsigned space = 0;

	space += si_descriptor_list_cs_space(SI_NUM_SHADER_BUFFERS +
					     SI_NUM_CONST_BUFFERS, 4);
	/* two 8-byte images share one 16-byte slot */
	space += si_descriptor_list_cs_space(SI_NUM_IMAGES / 2 +
					     SI_NUM_SAMPLERS, 16);
	space *= SI_NUM_SHADERS;

	space += si_descriptor_list_cs_space(SI_NUM_RW_BUFFERS, 4);

	/* Increment CE counter packet */
	space += 2;

	return space;
}

/* initialize */
void si_need_cs_space(struct si_context *ctx)
{
	struct radeon_winsys_cs *cs = ctx->b.gfx.cs;
	struct radeon_winsys_cs *ce_ib = ctx->ce_ib;

	/* There is no need to flush the DMA IB here, because
	 * r600_need_dma_space always flushes the GFX IB if there is
	 * a conflict, which means any unflushed DMA commands automatically
	 * precede the GFX IB (= they had no dependency on the GFX IB when
	 * they were submitted).
	 */

	/* There are two memory usage counters in the winsys for all buffers
	 * that have been added (cs_add_buffer) and two counters in the pipe
	 * driver for those that haven't been added yet.
	 */
	if (unlikely(!radeon_cs_memory_below_limit(ctx->b.screen, ctx->b.gfx.cs,
						   ctx->b.vram, ctx->b.gtt))) {
		ctx->b.gtt = 0;
		ctx->b.vram = 0;
		ctx->b.gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
		return;
	}
	ctx->b.gtt = 0;
	ctx->b.vram = 0;

	/* If the CS is sufficiently large, don't count the space needed
	 * and just flush if there is not enough space left.
	 */
	if (!ctx->b.ws->cs_check_space(cs, 2048) ||
	    (ce_ib && !ctx->b.ws->cs_check_space(ce_ib, si_ce_needed_cs_space())))
		ctx->b.gfx.flush(ctx, RADEON_FLUSH_ASYNC, NULL);
}

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;

	if (!radeon_emitted(cs, ctx->b.initial_gfx_cs_size))
		return;

	if (r600_check_device_reset(&ctx->b))
		return;

	if (ctx->screen->b.debug_flags & DBG_CHECK_VM)
		flags &= ~RADEON_FLUSH_ASYNC;

	/* If the state tracker is flushing the GFX IB, r600_flush_from_st is
	 * responsible for flushing the DMA IB and merging the fences from both.
	 * This code is only needed when the driver flushes the GFX IB
	 * internally, and it never asks for a fence handle.
	 */
	if (radeon_emitted(ctx->b.dma.cs, 0)) {
		assert(fence == NULL); /* internal flushes only */
		ctx->b.dma.flush(ctx, flags, NULL);
	}

	ctx->gfx_flush_in_progress = true;

	/* This CE dump should be done in parallel with the last draw. */
	if (ctx->ce_ib)
		si_ce_save_all_descriptors_at_ib_end(ctx);

	r600_preflush_suspend_features(&ctx->b);

	ctx->b.flags |= SI_CONTEXT_CS_PARTIAL_FLUSH |
			SI_CONTEXT_PS_PARTIAL_FLUSH;

	/* DRM 3.1.0 doesn't flush TC for VI correctly. */
	if (ctx->b.chip_class == VI && ctx->b.screen->info.drm_minor <= 1)
		ctx->b.flags |= SI_CONTEXT_INV_GLOBAL_L2 |
				SI_CONTEXT_INV_VMEM_L1;

	si_emit_cache_flush(ctx);

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

	if (ctx->is_debug) {
		/* Save the IB for debug contexts. */
		radeon_clear_saved_cs(&ctx->last_gfx);
		radeon_save_cs(ws, cs, &ctx->last_gfx);
		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->b.last_gfx_fence);
	if (fence)
		ws->fence_reference(fence, ctx->b.last_gfx_fence);
	ctx->b.num_gfx_cs_flushes++;

	/* Check VM faults if needed. */
	if (ctx->screen->b.debug_flags & DBG_CHECK_VM) {
		/* Use conservative timeout 800ms, after which we won't wait any
		 * longer and assume the GPU is hung.
		 */
		ctx->b.ws->fence_wait(ctx->b.ws, ctx->b.last_gfx_fence, 800*1000*1000);

		si_check_vm_faults(&ctx->b, &ctx->last_gfx, RING_GFX);
	}

	si_begin_new_cs(ctx);
	ctx->gfx_flush_in_progress = false;
}

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, 0,
						    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 not flushed by the kernel. */
	if (ctx->b.chip_class >= CIK)
		ctx->b.flags |= SI_CONTEXT_INV_SMEM_L1 |
				SI_CONTEXT_INV_ICACHE;

	ctx->b.flags |= R600_CONTEXT_START_PIPELINE_STATS;

	/* 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);

	if (ctx->ce_preamble_ib)
		si_ce_enable_loads(ctx->ce_preamble_ib);
	else if (ctx->ce_ib)
		si_ce_enable_loads(ctx->ce_ib);

	if (ctx->ce_ib)
		si_ce_restore_all_descriptors_at_ib_start(ctx);

	if (ctx->b.chip_class >= CIK)
		si_mark_atom_dirty(ctx, &ctx->prefetch_L2);

	/* CLEAR_STATE disables all colorbuffers, so only enable bound ones. */
	ctx->framebuffer.dirty_cbufs =
		u_bit_consecutive(0, ctx->framebuffer.state.nr_cbufs);
	/* CLEAR_STATE disables the zbuffer, so only enable it if it's bound. */
	ctx->framebuffer.dirty_zsbuf = ctx->framebuffer.state.zsbuf != NULL;
	/* This should always be marked as dirty to set the framebuffer scissor
	 * at least. */
	si_mark_atom_dirty(ctx, &ctx->framebuffer.atom);

	si_mark_atom_dirty(ctx, &ctx->clip_regs);
	/* CLEAR_STATE sets zeros. */
	if (ctx->clip_state.any_nonzeros)
		si_mark_atom_dirty(ctx, &ctx->clip_state.atom);
	ctx->msaa_sample_locs.nr_samples = 0;
	si_mark_atom_dirty(ctx, &ctx->msaa_sample_locs.atom);
	si_mark_atom_dirty(ctx, &ctx->msaa_config);
	/* CLEAR_STATE sets 0xffff. */
	if (ctx->sample_mask.sample_mask != 0xffff)
		si_mark_atom_dirty(ctx, &ctx->sample_mask.atom);
	si_mark_atom_dirty(ctx, &ctx->cb_render_state);
	/* CLEAR_STATE sets zeros. */
	if (ctx->blend_color.any_nonzeros)
		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);
	si_all_resident_buffers_begin_new_cs(ctx);

	ctx->b.scissors.dirty_mask = (1 << R600_MAX_VIEWPORTS) - 1;
	ctx->b.viewports.dirty_mask = (1 << R600_MAX_VIEWPORTS) - 1;
	ctx->b.viewports.depth_range_dirty_mask = (1 << R600_MAX_VIEWPORTS) - 1;
	si_mark_atom_dirty(ctx, &ctx->b.scissors.atom);
	si_mark_atom_dirty(ctx, &ctx->b.viewports.atom);

	si_mark_atom_dirty(ctx, &ctx->scratch_state);
	if (ctx->scratch_buffer) {
		r600_context_add_resource_size(&ctx->b.b,
					       &ctx->scratch_buffer->b.b);
	}

	r600_postflush_resume_features(&ctx->b);

	assert(!ctx->b.gfx.cs->prev_dw);
	ctx->b.initial_gfx_cs_size = ctx->b.gfx.cs->current.cdw;

	/* Invalidate various draw states so that they are emitted before
	 * the first draw call. */
	si_invalidate_draw_sh_constants(ctx);
	ctx->last_index_size = -1;
	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_rast_prim = -1;
	ctx->last_sc_line_stipple = ~0;
	ctx->last_vs_state = ~0;
	ctx->last_ls = NULL;
	ctx->last_tcs = NULL;
	ctx->last_tes_sh_base = -1;
	ctx->last_num_tcs_input_cp = -1;

	ctx->cs_shader_state.initialized = false;
}