freedreno: don't overflow cmdstream buffer so much
[mesa.git] / src / gallium / drivers / freedreno / freedreno_draw.c
1 /* -*- mode: C; c-file-style: "k&r"; tab-width 4; indent-tabs-mode: t; -*- */
2
3 /*
4 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
5 *
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
12 *
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
15 * Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
23 * SOFTWARE.
24 *
25 * Authors:
26 * Rob Clark <robclark@freedesktop.org>
27 */
28
29 #include "pipe/p_state.h"
30 #include "util/u_string.h"
31 #include "util/u_memory.h"
32 #include "util/u_prim.h"
33 #include "util/u_format.h"
34
35 #include "freedreno_draw.h"
36 #include "freedreno_context.h"
37 #include "freedreno_state.h"
38 #include "freedreno_resource.h"
39 #include "freedreno_query_hw.h"
40 #include "freedreno_util.h"
41
42
43 static enum pc_di_index_size
44 size2indextype(unsigned index_size)
45 {
46 switch (index_size) {
47 case 1: return INDEX_SIZE_8_BIT;
48 case 2: return INDEX_SIZE_16_BIT;
49 case 4: return INDEX_SIZE_32_BIT;
50 }
51 DBG("unsupported index size: %d", index_size);
52 assert(0);
53 return INDEX_SIZE_IGN;
54 }
55
56 /* this is same for a2xx/a3xx, so split into helper: */
57 void
58 fd_draw_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
59 enum pc_di_vis_cull_mode vismode,
60 const struct pipe_draw_info *info)
61 {
62 struct pipe_index_buffer *idx = &ctx->indexbuf;
63 struct fd_bo *idx_bo = NULL;
64 enum pc_di_index_size idx_type = INDEX_SIZE_IGN;
65 enum pc_di_src_sel src_sel;
66 uint32_t idx_size, idx_offset;
67
68 if (info->indexed) {
69 assert(!idx->user_buffer);
70
71 idx_bo = fd_resource(idx->buffer)->bo;
72 idx_type = size2indextype(idx->index_size);
73 idx_size = idx->index_size * info->count;
74 idx_offset = idx->offset + (info->start * idx->index_size);
75 src_sel = DI_SRC_SEL_DMA;
76 } else {
77 idx_bo = NULL;
78 idx_type = INDEX_SIZE_IGN;
79 idx_size = 0;
80 idx_offset = 0;
81 src_sel = DI_SRC_SEL_AUTO_INDEX;
82 }
83
84 fd_draw(ctx, ring, ctx->primtypes[info->mode], vismode, src_sel,
85 info->count, idx_type, idx_size, idx_offset, idx_bo);
86 }
87
88 static void
89 fd_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
90 {
91 struct fd_context *ctx = fd_context(pctx);
92 struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
93 struct pipe_scissor_state *scissor = fd_context_get_scissor(ctx);
94 unsigned i, buffers = 0;
95
96 /* if we supported transform feedback, we'd have to disable this: */
97 if (((scissor->maxx - scissor->minx) *
98 (scissor->maxy - scissor->miny)) == 0) {
99 return;
100 }
101
102 /* emulate unsupported primitives: */
103 if (!fd_supported_prim(ctx, info->mode)) {
104 util_primconvert_save_index_buffer(ctx->primconvert, &ctx->indexbuf);
105 util_primconvert_save_rasterizer_state(ctx->primconvert, ctx->rasterizer);
106 util_primconvert_draw_vbo(ctx->primconvert, info);
107 return;
108 }
109
110 ctx->needs_flush = true;
111
112 /*
113 * Figure out the buffers/features we need:
114 */
115
116 if (fd_depth_enabled(ctx)) {
117 buffers |= FD_BUFFER_DEPTH;
118 fd_resource(pfb->zsbuf->texture)->dirty = true;
119 ctx->gmem_reason |= FD_GMEM_DEPTH_ENABLED;
120 }
121
122 if (fd_stencil_enabled(ctx)) {
123 buffers |= FD_BUFFER_STENCIL;
124 fd_resource(pfb->zsbuf->texture)->dirty = true;
125 ctx->gmem_reason |= FD_GMEM_STENCIL_ENABLED;
126 }
127
128 if (fd_logicop_enabled(ctx))
129 ctx->gmem_reason |= FD_GMEM_LOGICOP_ENABLED;
130
131 for (i = 0; i < pfb->nr_cbufs; i++) {
132 struct pipe_resource *surf;
133
134 if (!pfb->cbufs[i])
135 continue;
136
137 surf = pfb->cbufs[i]->texture;
138
139 fd_resource(surf)->dirty = true;
140 buffers |= FD_BUFFER_COLOR;
141
142 if (surf->nr_samples > 1)
143 ctx->gmem_reason |= FD_GMEM_MSAA_ENABLED;
144
145 if (fd_blend_enabled(ctx, i))
146 ctx->gmem_reason |= FD_GMEM_BLEND_ENABLED;
147 }
148
149 ctx->num_draws++;
150
151 ctx->stats.draw_calls++;
152 ctx->stats.prims_emitted +=
153 u_reduced_prims_for_vertices(info->mode, info->count);
154
155 /* any buffers that haven't been cleared, we need to restore: */
156 ctx->restore |= buffers & (FD_BUFFER_ALL & ~ctx->cleared);
157 /* and any buffers used, need to be resolved: */
158 ctx->resolve |= buffers;
159
160 DBG("%x num_draws=%u (%s/%s)", buffers, ctx->num_draws,
161 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
162 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
163
164 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_DRAW);
165 ctx->draw(ctx, info);
166
167 /* if an app (or, well, piglit test) does many thousands of draws
168 * without flush (or anything which implicitly flushes, like
169 * changing render targets), we can exceed the ringbuffer size.
170 * Since we don't currently have a sane way to wrapparound, and
171 * we use the same buffer for both draw and tiling commands, for
172 * now we need to do this hack and trigger flush if we are running
173 * low on remaining space for cmds:
174 */
175 if ((ctx->ring->cur - ctx->ring->start) > ctx->ring->size/8)
176 fd_context_render(pctx);
177 }
178
179 /* TODO figure out how to make better use of existing state mechanism
180 * for clear (and possibly gmem->mem / mem->gmem) so we can (a) keep
181 * track of what state really actually changes, and (b) reduce the code
182 * in the a2xx/a3xx parts.
183 */
184
185 static void
186 fd_clear(struct pipe_context *pctx, unsigned buffers,
187 const union pipe_color_union *color, double depth, unsigned stencil)
188 {
189 struct fd_context *ctx = fd_context(pctx);
190 struct pipe_framebuffer_state *pfb = &ctx->framebuffer;
191
192 ctx->cleared |= buffers;
193 ctx->resolve |= buffers;
194 ctx->needs_flush = true;
195
196 if (buffers & PIPE_CLEAR_COLOR)
197 fd_resource(pfb->cbufs[0]->texture)->dirty = true;
198
199 if (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
200 fd_resource(pfb->zsbuf->texture)->dirty = true;
201 ctx->gmem_reason |= FD_GMEM_CLEARS_DEPTH_STENCIL;
202 }
203
204 DBG("%x depth=%f, stencil=%u (%s/%s)", buffers, depth, stencil,
205 util_format_short_name(pipe_surface_format(pfb->cbufs[0])),
206 util_format_short_name(pipe_surface_format(pfb->zsbuf)));
207
208 fd_hw_query_set_stage(ctx, ctx->ring, FD_STAGE_CLEAR);
209
210 ctx->clear(ctx, buffers, color, depth, stencil);
211
212 ctx->dirty |= FD_DIRTY_ZSA |
213 FD_DIRTY_VIEWPORT |
214 FD_DIRTY_RASTERIZER |
215 FD_DIRTY_SAMPLE_MASK |
216 FD_DIRTY_PROG |
217 FD_DIRTY_CONSTBUF |
218 FD_DIRTY_BLEND;
219
220 if (fd_mesa_debug & FD_DBG_DCLEAR)
221 ctx->dirty = 0xffffffff;
222 }
223
224 static void
225 fd_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
226 const union pipe_color_union *color,
227 unsigned x, unsigned y, unsigned w, unsigned h)
228 {
229 DBG("TODO: x=%u, y=%u, w=%u, h=%u", x, y, w, h);
230 }
231
232 static void
233 fd_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
234 unsigned buffers, double depth, unsigned stencil,
235 unsigned x, unsigned y, unsigned w, unsigned h)
236 {
237 DBG("TODO: buffers=%u, depth=%f, stencil=%u, x=%u, y=%u, w=%u, h=%u",
238 buffers, depth, stencil, x, y, w, h);
239 }
240
241 void
242 fd_draw_init(struct pipe_context *pctx)
243 {
244 pctx->draw_vbo = fd_draw_vbo;
245 pctx->clear = fd_clear;
246 pctx->clear_render_target = fd_clear_render_target;
247 pctx->clear_depth_stencil = fd_clear_depth_stencil;
248 }