tu_util.c revision 7ec681f3 
1/*
2 * Copyright © 2015 Intel Corporation
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
20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21 * DEALINGS IN THE SOFTWARE.
22 */
23
24#include "tu_private.h"
25
26#include <assert.h>
27#include <errno.h>
28#include <stdarg.h>
29#include <stdio.h>
30#include <stdlib.h>
31#include <string.h>
32
33#include "util/u_math.h"
34#include "vk_enum_to_str.h"
35
36void PRINTFLIKE(3, 4)
37   __tu_finishme(const char *file, int line, const char *format, ...)
38{
39   va_list ap;
40   char buffer[256];
41
42   va_start(ap, format);
43   vsnprintf(buffer, sizeof(buffer), format, ap);
44   va_end(ap);
45
46   mesa_loge("%s:%d: FINISHME: %s\n", file, line, buffer);
47}
48
49VkResult
50__vk_startup_errorf(struct tu_instance *instance,
51                    VkResult error,
52                    bool always_print,
53                    const char *file,
54                    int line,
55                    const char *format,
56                    ...)
57{
58   va_list ap;
59   char buffer[256];
60
61   const char *error_str = vk_Result_to_str(error);
62
63#ifndef DEBUG
64   if (!always_print)
65      return error;
66#endif
67
68   if (format) {
69      va_start(ap, format);
70      vsnprintf(buffer, sizeof(buffer), format, ap);
71      va_end(ap);
72
73      mesa_loge("%s:%d: %s (%s)\n", file, line, buffer, error_str);
74   } else {
75      mesa_loge("%s:%d: %s\n", file, line, error_str);
76   }
77
78   return error;
79}
80
81static void
82tu_tiling_config_update_tile_layout(struct tu_framebuffer *fb,
83                                    const struct tu_device *dev,
84                                    const struct tu_render_pass *pass)
85{
86   const uint32_t tile_align_w = pass->tile_align_w;
87   const uint32_t tile_align_h = dev->physical_device->info->tile_align_h;
88   const uint32_t max_tile_width = dev->physical_device->info->tile_max_w;
89   const uint32_t max_tile_height = dev->physical_device->info->tile_max_h;
90
91   /* start from 1 tile */
92   fb->tile_count = (VkExtent2D) {
93      .width = 1,
94      .height = 1,
95   };
96   fb->tile0 = (VkExtent2D) {
97      .width = util_align_npot(fb->width, tile_align_w),
98      .height = align(fb->height, tile_align_h),
99   };
100
101   /* will force to sysmem, don't bother trying to have a valid tile config
102    * TODO: just skip all GMEM stuff when sysmem is forced?
103    */
104   if (!pass->gmem_pixels)
105      return;
106
107   if (unlikely(dev->physical_device->instance->debug_flags & TU_DEBUG_FORCEBIN)) {
108      /* start with 2x2 tiles */
109      fb->tile_count.width = 2;
110      fb->tile_count.height = 2;
111      fb->tile0.width = util_align_npot(DIV_ROUND_UP(fb->width, 2), tile_align_w);
112      fb->tile0.height = align(DIV_ROUND_UP(fb->height, 2), tile_align_h);
113   }
114
115   /* do not exceed max tile width */
116   while (fb->tile0.width > max_tile_width) {
117      fb->tile_count.width++;
118      fb->tile0.width =
119         util_align_npot(DIV_ROUND_UP(fb->width, fb->tile_count.width), tile_align_w);
120   }
121
122   /* do not exceed max tile height */
123   while (fb->tile0.height > max_tile_height) {
124      fb->tile_count.height++;
125      fb->tile0.height =
126         util_align_npot(DIV_ROUND_UP(fb->height, fb->tile_count.height), tile_align_h);
127   }
128
129   /* do not exceed gmem size */
130   while (fb->tile0.width * fb->tile0.height > pass->gmem_pixels) {
131      if (fb->tile0.width > MAX2(tile_align_w, fb->tile0.height)) {
132         fb->tile_count.width++;
133         fb->tile0.width =
134            util_align_npot(DIV_ROUND_UP(fb->width, fb->tile_count.width), tile_align_w);
135      } else {
136         /* if this assert fails then layout is impossible.. */
137         assert(fb->tile0.height > tile_align_h);
138         fb->tile_count.height++;
139         fb->tile0.height =
140            align(DIV_ROUND_UP(fb->height, fb->tile_count.height), tile_align_h);
141      }
142   }
143}
144
145static void
146tu_tiling_config_update_pipe_layout(struct tu_framebuffer *fb,
147                                    const struct tu_device *dev)
148{
149   const uint32_t max_pipe_count = 32; /* A6xx */
150
151   /* start from 1 tile per pipe */
152   fb->pipe0 = (VkExtent2D) {
153      .width = 1,
154      .height = 1,
155   };
156   fb->pipe_count = fb->tile_count;
157
158   while (fb->pipe_count.width * fb->pipe_count.height > max_pipe_count) {
159      if (fb->pipe0.width < fb->pipe0.height) {
160         fb->pipe0.width += 1;
161         fb->pipe_count.width =
162            DIV_ROUND_UP(fb->tile_count.width, fb->pipe0.width);
163      } else {
164         fb->pipe0.height += 1;
165         fb->pipe_count.height =
166            DIV_ROUND_UP(fb->tile_count.height, fb->pipe0.height);
167      }
168   }
169}
170
171static void
172tu_tiling_config_update_pipes(struct tu_framebuffer *fb,
173                              const struct tu_device *dev)
174{
175   const uint32_t max_pipe_count = 32; /* A6xx */
176   const uint32_t used_pipe_count =
177      fb->pipe_count.width * fb->pipe_count.height;
178   const VkExtent2D last_pipe = {
179      .width = (fb->tile_count.width - 1) % fb->pipe0.width + 1,
180      .height = (fb->tile_count.height - 1) % fb->pipe0.height + 1,
181   };
182
183   assert(used_pipe_count <= max_pipe_count);
184   assert(max_pipe_count <= ARRAY_SIZE(fb->pipe_config));
185
186   for (uint32_t y = 0; y < fb->pipe_count.height; y++) {
187      for (uint32_t x = 0; x < fb->pipe_count.width; x++) {
188         const uint32_t pipe_x = fb->pipe0.width * x;
189         const uint32_t pipe_y = fb->pipe0.height * y;
190         const uint32_t pipe_w = (x == fb->pipe_count.width - 1)
191                                    ? last_pipe.width
192                                    : fb->pipe0.width;
193         const uint32_t pipe_h = (y == fb->pipe_count.height - 1)
194                                    ? last_pipe.height
195                                    : fb->pipe0.height;
196         const uint32_t n = fb->pipe_count.width * y + x;
197
198         fb->pipe_config[n] = A6XX_VSC_PIPE_CONFIG_REG_X(pipe_x) |
199                                  A6XX_VSC_PIPE_CONFIG_REG_Y(pipe_y) |
200                                  A6XX_VSC_PIPE_CONFIG_REG_W(pipe_w) |
201                                  A6XX_VSC_PIPE_CONFIG_REG_H(pipe_h);
202         fb->pipe_sizes[n] = CP_SET_BIN_DATA5_0_VSC_SIZE(pipe_w * pipe_h);
203      }
204   }
205
206   memset(fb->pipe_config + used_pipe_count, 0,
207          sizeof(uint32_t) * (max_pipe_count - used_pipe_count));
208}
209
210void
211tu_framebuffer_tiling_config(struct tu_framebuffer *fb,
212                             const struct tu_device *device,
213                             const struct tu_render_pass *pass)
214{
215   tu_tiling_config_update_tile_layout(fb, device, pass);
216   tu_tiling_config_update_pipe_layout(fb, device);
217   tu_tiling_config_update_pipes(fb, device);
218}
219