1 /*
2  * Copyright © 2014 Broadcom
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 DEALINGS
21  * IN THE SOFTWARE.
22  */
23 
24 /** @file vc4_tiling.c
25  *
26  * Handles information about the VC4 tiling formats, and loading and storing
27  * from them.
28  *
29  * Texture mipmap levels on VC4 are (with the exception of 32-bit RGBA raster
30  * textures for scanout) stored as groups of microtiles.  If the texture is at
31  * least 4x4 microtiles (utiles), then those microtiles are arranged in a sort
32  * of Hilbert-fractal-ish layout (T), otherwise the microtiles are in raster
33  * order (LT).
34  *
35  * Specifically, the T format has:
36  *
37  * - 64b utiles of pixels in a raster-order grid according to cpp.  It's 4x4
38  *   pixels at 32 bit depth.
39  *
40  * - 1k subtiles made of a 4x4 raster-order grid of 64b utiles (so usually
41  *   16x16 pixels).
42  *
43  * - 4k tiles made of a 2x2 grid of 1k subtiles (so usually 32x32 pixels).  On
44  *   even 4k tile rows, they're arranged as (BL, TL, TR, BR), and on odd rows
45  *   they're (TR, BR, BL, TL), where bottom left is start of memory.
46  *
47  * - an image made of 4k tiles in rows either left-to-right (even rows of 4k
48  *   tiles) or right-to-left (odd rows of 4k tiles).
49  */
50 
51 #include "vc4_screen.h"
52 #include "vc4_context.h"
53 #include "vc4_tiling.h"
54 
55 /**
56  * The texture unit decides what tiling format a particular miplevel is using
57  * this function, so we lay out our miptrees accordingly.
58  */
59 bool
vc4_size_is_lt(uint32_t width,uint32_t height,int cpp)60 vc4_size_is_lt(uint32_t width, uint32_t height, int cpp)
61 {
62         return (width <= 4 * vc4_utile_width(cpp) ||
63                 height <= 4 * vc4_utile_height(cpp));
64 }
65 
66 /**
67  * Takes a utile x and y (and the number of utiles of width of the image) and
68  * returns the offset to the utile within a VC4_TILING_FORMAT_TF image.
69  */
70 static uint32_t
t_utile_address(uint32_t utile_x,uint32_t utile_y,uint32_t utile_stride)71 t_utile_address(uint32_t utile_x, uint32_t utile_y,
72                 uint32_t utile_stride)
73 {
74         /* T images have to be aligned to 8 utiles (4x4 subtiles, which are
75          * 2x2 in a 4k tile).
76          */
77         assert(!(utile_stride & 7));
78         uint32_t tile_stride = utile_stride >> 3;
79         /* 4k tile offsets. */
80         uint32_t tile_x = utile_x >> 3;
81         uint32_t tile_y = utile_y >> 3;
82         bool odd_tile_y = tile_y & 1;
83 
84         /* Odd lines of 4k tiles go right-to-left. */
85         if (odd_tile_y)
86                 tile_x = tile_stride - tile_x - 1;
87 
88         uint32_t tile_offset = 4096 * (tile_y * tile_stride + tile_x);
89 
90         uint32_t stile_x = (utile_x >> 2) & 1;
91         uint32_t stile_y = (utile_y >> 2) & 1;
92         uint32_t stile_index = (stile_y << 1) + stile_x;
93         static const uint32_t odd_stile_map[4] = {2, 1, 3, 0};
94         static const uint32_t even_stile_map[4] = {0, 3, 1, 2};
95 
96         uint32_t stile_offset = 1024 * (odd_tile_y ?
97                                         odd_stile_map[stile_index] :
98                                         even_stile_map[stile_index]);
99 
100         /* This function no longer handles the utile offset within a subtile.
101          * Walking subtiles is the job of the LT image handler.
102          */
103         assert(!(utile_x & 3) && !(utile_y & 3));
104 
105 #if 0
106         fprintf(stderr, "utile %d,%d -> %d + %d + %d (stride %d,%d) = %d\n",
107                 utile_x, utile_y,
108                 tile_offset, stile_offset, utile_offset,
109                 utile_stride, tile_stride,
110                 tile_offset + stile_offset + utile_offset);
111 #endif
112 
113         return tile_offset + stile_offset;
114 }
115 
116 /**
117  * Loads or stores a T texture image by breaking it down into subtiles
118  * (1024-byte, 4x4-utile) sub-images that we can use the LT tiling functions
119  * on.
120  */
121 static inline void
vc4_t_image_helper(void * gpu,uint32_t gpu_stride,void * cpu,uint32_t cpu_stride,int cpp,const struct pipe_box * box,bool to_cpu)122 vc4_t_image_helper(void *gpu, uint32_t gpu_stride,
123                    void *cpu, uint32_t cpu_stride,
124                    int cpp, const struct pipe_box *box,
125                    bool to_cpu)
126 {
127         uint32_t utile_w = vc4_utile_width(cpp);
128         uint32_t utile_h = vc4_utile_height(cpp);
129         uint32_t utile_w_shift = ffs(utile_w) - 1;
130         uint32_t utile_h_shift = ffs(utile_h) - 1;
131         uint32_t stile_w = 4 * utile_w;
132         uint32_t stile_h = 4 * utile_h;
133         assert(stile_w * stile_h * cpp == 1024);
134         uint32_t utile_stride = gpu_stride / cpp / utile_w;
135         uint32_t x1 = box->x;
136         uint32_t y1 = box->y;
137         uint32_t x2 = box->x + box->width;
138         uint32_t y2 = box->y + box->height;
139         struct pipe_box partial_box;
140         uint32_t gpu_lt_stride = stile_w * cpp;
141 
142         for (uint32_t y = y1; y < y2; y = align(y + 1, stile_h)) {
143                 partial_box.y = y & (stile_h - 1);
144                 partial_box.height = MIN2(y2 - y, stile_h - partial_box.y);
145 
146                 uint32_t cpu_offset = 0;
147                 for (uint32_t x = x1; x < x2; x = align(x + 1, stile_w)) {
148                         partial_box.x = x & (stile_w - 1);
149                         partial_box.width = MIN2(x2 - x,
150                                                  stile_w - partial_box.x);
151 
152                         /* The dst offset we want is the start of this
153                          * subtile
154                          */
155                         uint32_t gpu_offset =
156                                 t_utile_address((x >> utile_w_shift) & ~0x3,
157                                                 (y >> utile_h_shift) & ~0x3,
158                                                 utile_stride);
159 
160                         if (to_cpu) {
161                                 vc4_load_lt_image(cpu + cpu_offset,
162                                                   cpu_stride,
163                                                   gpu + gpu_offset,
164                                                   gpu_lt_stride,
165                                                   cpp, &partial_box);
166                         } else {
167                                 vc4_store_lt_image(gpu + gpu_offset,
168                                                    gpu_lt_stride,
169                                                    cpu + cpu_offset,
170                                                    cpu_stride,
171                                                    cpp, &partial_box);
172                         }
173 
174                         cpu_offset += partial_box.width * cpp;
175                 }
176                 cpu += cpu_stride * partial_box.height;
177         }
178 }
179 
180 static void
vc4_store_t_image(void * dst,uint32_t dst_stride,void * src,uint32_t src_stride,int cpp,const struct pipe_box * box)181 vc4_store_t_image(void *dst, uint32_t dst_stride,
182                   void *src, uint32_t src_stride,
183                   int cpp, const struct pipe_box *box)
184 {
185         vc4_t_image_helper(dst, dst_stride,
186                            src, src_stride,
187                            cpp, box, false);
188 }
189 
190 static void
vc4_load_t_image(void * dst,uint32_t dst_stride,void * src,uint32_t src_stride,int cpp,const struct pipe_box * box)191 vc4_load_t_image(void *dst, uint32_t dst_stride,
192                   void *src, uint32_t src_stride,
193                   int cpp, const struct pipe_box *box)
194 {
195         vc4_t_image_helper(src, src_stride,
196                            dst, dst_stride,
197                            cpp, box, true);
198 }
199 
200 /**
201  * Loads pixel data from the start (microtile-aligned) box in \p src to the
202  * start of \p dst according to the given tiling format.
203  */
204 void
vc4_load_tiled_image(void * dst,uint32_t dst_stride,void * src,uint32_t src_stride,uint8_t tiling_format,int cpp,const struct pipe_box * box)205 vc4_load_tiled_image(void *dst, uint32_t dst_stride,
206                      void *src, uint32_t src_stride,
207                      uint8_t tiling_format, int cpp,
208                      const struct pipe_box *box)
209 {
210         if (tiling_format == VC4_TILING_FORMAT_LT) {
211                 vc4_load_lt_image(dst, dst_stride,
212                                   src, src_stride,
213                                   cpp, box);
214         } else {
215                 assert(tiling_format == VC4_TILING_FORMAT_T);
216                 vc4_load_t_image(dst, dst_stride,
217                                  src, src_stride,
218                                  cpp, box);
219         }
220 }
221 
222 /**
223  * Stores pixel data from the start of \p src into a (microtile-aligned) box in
224  * \p dst according to the given tiling format.
225  */
226 void
vc4_store_tiled_image(void * dst,uint32_t dst_stride,void * src,uint32_t src_stride,uint8_t tiling_format,int cpp,const struct pipe_box * box)227 vc4_store_tiled_image(void *dst, uint32_t dst_stride,
228                       void *src, uint32_t src_stride,
229                       uint8_t tiling_format, int cpp,
230                       const struct pipe_box *box)
231 {
232         if (tiling_format == VC4_TILING_FORMAT_LT) {
233                 vc4_store_lt_image(dst, dst_stride,
234                                    src, src_stride,
235                                    cpp, box);
236         } else {
237                 assert(tiling_format == VC4_TILING_FORMAT_T);
238                 vc4_store_t_image(dst, dst_stride,
239                                   src, src_stride,
240                                   cpp, box);
241         }
242 }
243 
244