freedreno/a5xx: fallback to slow-clear for z32
[mesa.git] / src / gallium / drivers / freedreno / a5xx / fd5_draw.c
1 /*
2 * Copyright (C) 2016 Rob Clark <robclark@freedesktop.org>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 * SOFTWARE.
22 *
23 * Authors:
24 * Rob Clark <robclark@freedesktop.org>
25 */
26
27 #include "pipe/p_state.h"
28 #include "util/u_string.h"
29 #include "util/u_memory.h"
30 #include "util/u_prim.h"
31
32 #include "freedreno_state.h"
33 #include "freedreno_resource.h"
34
35 #include "fd5_draw.h"
36 #include "fd5_context.h"
37 #include "fd5_emit.h"
38 #include "fd5_program.h"
39 #include "fd5_format.h"
40 #include "fd5_zsa.h"
41
42
43 static void
44 draw_impl(struct fd_context *ctx, struct fd_ringbuffer *ring,
45 struct fd5_emit *emit, unsigned index_offset)
46 {
47 const struct pipe_draw_info *info = emit->info;
48 enum pc_di_primtype primtype = ctx->primtypes[info->mode];
49
50 fd5_emit_state(ctx, ring, emit);
51
52 if (emit->dirty & (FD_DIRTY_VTXBUF | FD_DIRTY_VTXSTATE))
53 fd5_emit_vertex_bufs(ring, emit);
54
55 OUT_PKT4(ring, REG_A5XX_VFD_INDEX_OFFSET, 2);
56 OUT_RING(ring, info->index_size ? info->index_bias : info->start); /* VFD_INDEX_OFFSET */
57 OUT_RING(ring, info->start_instance); /* ??? UNKNOWN_2209 */
58
59 OUT_PKT4(ring, REG_A5XX_PC_RESTART_INDEX, 1);
60 OUT_RING(ring, info->primitive_restart ? /* PC_RESTART_INDEX */
61 info->restart_index : 0xffffffff);
62
63 fd5_emit_render_cntl(ctx, false, emit->key.binning_pass);
64 fd5_draw_emit(ctx->batch, ring, primtype,
65 emit->key.binning_pass ? IGNORE_VISIBILITY : USE_VISIBILITY,
66 info, index_offset);
67 }
68
69 /* fixup dirty shader state in case some "unrelated" (from the state-
70 * tracker's perspective) state change causes us to switch to a
71 * different variant.
72 */
73 static void
74 fixup_shader_state(struct fd_context *ctx, struct ir3_shader_key *key)
75 {
76 struct fd5_context *fd5_ctx = fd5_context(ctx);
77 struct ir3_shader_key *last_key = &fd5_ctx->last_key;
78
79 if (!ir3_shader_key_equal(last_key, key)) {
80 if (ir3_shader_key_changes_fs(last_key, key)) {
81 ctx->dirty_shader[PIPE_SHADER_FRAGMENT] |= FD_DIRTY_SHADER_PROG;
82 ctx->dirty |= FD_DIRTY_PROG;
83 }
84
85 if (ir3_shader_key_changes_vs(last_key, key)) {
86 ctx->dirty_shader[PIPE_SHADER_VERTEX] |= FD_DIRTY_SHADER_PROG;
87 ctx->dirty |= FD_DIRTY_PROG;
88 }
89
90 fd5_ctx->last_key = *key;
91 }
92 }
93
94 static bool
95 fd5_draw_vbo(struct fd_context *ctx, const struct pipe_draw_info *info,
96 unsigned index_offset)
97 {
98 struct fd5_context *fd5_ctx = fd5_context(ctx);
99 struct fd5_emit emit = {
100 .debug = &ctx->debug,
101 .vtx = &ctx->vtx,
102 .prog = &ctx->prog,
103 .info = info,
104 .key = {
105 .color_two_side = ctx->rasterizer->light_twoside,
106 .vclamp_color = ctx->rasterizer->clamp_vertex_color,
107 .fclamp_color = ctx->rasterizer->clamp_fragment_color,
108 .rasterflat = ctx->rasterizer->flatshade,
109 .half_precision = ctx->in_blit &&
110 fd_half_precision(&ctx->batch->framebuffer),
111 .ucp_enables = ctx->rasterizer->clip_plane_enable,
112 .has_per_samp = (fd5_ctx->fsaturate || fd5_ctx->vsaturate ||
113 fd5_ctx->fastc_srgb || fd5_ctx->vastc_srgb),
114 .vsaturate_s = fd5_ctx->vsaturate_s,
115 .vsaturate_t = fd5_ctx->vsaturate_t,
116 .vsaturate_r = fd5_ctx->vsaturate_r,
117 .fsaturate_s = fd5_ctx->fsaturate_s,
118 .fsaturate_t = fd5_ctx->fsaturate_t,
119 .fsaturate_r = fd5_ctx->fsaturate_r,
120 .vastc_srgb = fd5_ctx->vastc_srgb,
121 .fastc_srgb = fd5_ctx->fastc_srgb,
122 },
123 .rasterflat = ctx->rasterizer->flatshade,
124 .sprite_coord_enable = ctx->rasterizer->sprite_coord_enable,
125 .sprite_coord_mode = ctx->rasterizer->sprite_coord_mode,
126 };
127
128 fixup_shader_state(ctx, &emit.key);
129
130 unsigned dirty = ctx->dirty;
131
132 /* do regular pass first, since that is more likely to fail compiling: */
133
134 if (!(fd5_emit_get_vp(&emit) && fd5_emit_get_fp(&emit)))
135 return false;
136
137 emit.key.binning_pass = false;
138 emit.dirty = dirty;
139
140 draw_impl(ctx, ctx->batch->draw, &emit, index_offset);
141
142 /* and now binning pass: */
143 emit.key.binning_pass = true;
144 emit.dirty = dirty & ~(FD_DIRTY_BLEND);
145 emit.vp = NULL; /* we changed key so need to refetch vp */
146 emit.fp = NULL;
147 draw_impl(ctx, ctx->batch->binning, &emit, index_offset);
148
149 if (emit.streamout_mask) {
150 struct fd_ringbuffer *ring = ctx->batch->draw;
151
152 for (unsigned i = 0; i < PIPE_MAX_SO_BUFFERS; i++) {
153 if (emit.streamout_mask & (1 << i)) {
154 OUT_PKT7(ring, CP_EVENT_WRITE, 1);
155 OUT_RING(ring, FLUSH_SO_0 + i);
156 }
157 }
158 }
159
160 fd_context_all_clean(ctx);
161
162 return true;
163 }
164
165 static bool is_z32(enum pipe_format format)
166 {
167 switch (format) {
168 case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
169 case PIPE_FORMAT_Z32_UNORM:
170 case PIPE_FORMAT_Z32_FLOAT:
171 return true;
172 default:
173 return false;
174 }
175 }
176
177 static bool
178 fd5_clear(struct fd_context *ctx, unsigned buffers,
179 const union pipe_color_union *color, double depth, unsigned stencil)
180 {
181 struct fd_ringbuffer *ring = ctx->batch->draw;
182 struct pipe_framebuffer_state *pfb = &ctx->batch->framebuffer;
183 struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
184
185 if ((buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
186 is_z32(pfb->zsbuf->format))
187 return false;
188
189 /* TODO handle scissor.. or fallback to slow-clear? */
190
191 ctx->batch->max_scissor.minx = MIN2(ctx->batch->max_scissor.minx, scissor->minx);
192 ctx->batch->max_scissor.miny = MIN2(ctx->batch->max_scissor.miny, scissor->miny);
193 ctx->batch->max_scissor.maxx = MAX2(ctx->batch->max_scissor.maxx, scissor->maxx);
194 ctx->batch->max_scissor.maxy = MAX2(ctx->batch->max_scissor.maxy, scissor->maxy);
195
196 fd5_emit_render_cntl(ctx, true, false);
197
198 if (buffers & PIPE_CLEAR_COLOR) {
199 for (int i = 0; i < pfb->nr_cbufs; i++) {
200 union util_color uc = {0};
201
202 if (!pfb->cbufs[i])
203 continue;
204
205 if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
206 continue;
207
208 enum pipe_format pfmt = pfb->cbufs[i]->format;
209
210 // XXX I think RB_CLEAR_COLOR_DWn wants to take into account SWAP??
211 union pipe_color_union swapped;
212 switch (fd5_pipe2swap(pfmt)) {
213 case WZYX:
214 swapped.ui[0] = color->ui[0];
215 swapped.ui[1] = color->ui[1];
216 swapped.ui[2] = color->ui[2];
217 swapped.ui[3] = color->ui[3];
218 break;
219 case WXYZ:
220 swapped.ui[2] = color->ui[0];
221 swapped.ui[1] = color->ui[1];
222 swapped.ui[0] = color->ui[2];
223 swapped.ui[3] = color->ui[3];
224 break;
225 case ZYXW:
226 swapped.ui[3] = color->ui[0];
227 swapped.ui[0] = color->ui[1];
228 swapped.ui[1] = color->ui[2];
229 swapped.ui[2] = color->ui[3];
230 break;
231 case XYZW:
232 swapped.ui[3] = color->ui[0];
233 swapped.ui[2] = color->ui[1];
234 swapped.ui[1] = color->ui[2];
235 swapped.ui[0] = color->ui[3];
236 break;
237 }
238
239 if (util_format_is_pure_uint(pfmt)) {
240 util_format_write_4ui(pfmt, swapped.ui, 0, &uc, 0, 0, 0, 1, 1);
241 } else if (util_format_is_pure_sint(pfmt)) {
242 util_format_write_4i(pfmt, swapped.i, 0, &uc, 0, 0, 0, 1, 1);
243 } else {
244 util_pack_color(swapped.f, pfmt, &uc);
245 }
246
247 OUT_PKT4(ring, REG_A5XX_RB_BLIT_CNTL, 1);
248 OUT_RING(ring, A5XX_RB_BLIT_CNTL_BUF(BLIT_MRT0 + i));
249
250 OUT_PKT4(ring, REG_A5XX_RB_CLEAR_CNTL, 1);
251 OUT_RING(ring, A5XX_RB_CLEAR_CNTL_FAST_CLEAR |
252 A5XX_RB_CLEAR_CNTL_MASK(0xf));
253
254 OUT_PKT4(ring, REG_A5XX_RB_CLEAR_COLOR_DW0, 4);
255 OUT_RING(ring, uc.ui[0]); /* RB_CLEAR_COLOR_DW0 */
256 OUT_RING(ring, uc.ui[1]); /* RB_CLEAR_COLOR_DW1 */
257 OUT_RING(ring, uc.ui[2]); /* RB_CLEAR_COLOR_DW2 */
258 OUT_RING(ring, uc.ui[3]); /* RB_CLEAR_COLOR_DW3 */
259
260 fd5_emit_blit(ctx, ring);
261 }
262 }
263
264 if (pfb->zsbuf && (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
265 uint32_t clear =
266 util_pack_z_stencil(pfb->zsbuf->format, depth, stencil);
267 uint32_t mask = 0;
268
269 if (buffers & PIPE_CLEAR_DEPTH)
270 mask |= 0x1;
271
272 if (buffers & PIPE_CLEAR_STENCIL)
273 mask |= 0x2;
274
275 OUT_PKT4(ring, REG_A5XX_RB_BLIT_CNTL, 1);
276 OUT_RING(ring, A5XX_RB_BLIT_CNTL_BUF(BLIT_ZS));
277
278 OUT_PKT4(ring, REG_A5XX_RB_CLEAR_CNTL, 1);
279 OUT_RING(ring, A5XX_RB_CLEAR_CNTL_FAST_CLEAR |
280 A5XX_RB_CLEAR_CNTL_MASK(mask));
281
282 OUT_PKT4(ring, REG_A5XX_RB_CLEAR_COLOR_DW0, 1);
283 OUT_RING(ring, clear); /* RB_CLEAR_COLOR_DW0 */
284
285 fd5_emit_blit(ctx, ring);
286 }
287
288 /* disable fast clear to not interfere w/ gmem->mem, etc.. */
289 OUT_PKT4(ring, REG_A5XX_RB_CLEAR_CNTL, 1);
290 OUT_RING(ring, 0x00000000); /* RB_CLEAR_CNTL */
291
292 return true;
293 }
294
295 void
296 fd5_draw_init(struct pipe_context *pctx)
297 {
298 struct fd_context *ctx = fd_context(pctx);
299 ctx->draw_vbo = fd5_draw_vbo;
300 ctx->clear = fd5_clear;
301 }