0a80aefb80e0f5046f62f53c786cc35046d88ffb
[mesa.git] / src / freedreno / vulkan / tu_meta_copy.c
1 /*
2 * Copyright © 2016 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 "a6xx.xml.h"
27 #include "adreno_common.xml.h"
28 #include "adreno_pm4.xml.h"
29
30 #include "vk_format.h"
31
32 #include "tu_cs.h"
33
34 static uint32_t
35 blit_control(enum a6xx_color_fmt fmt)
36 {
37 unsigned blit_cntl = 0xf00000;
38 blit_cntl |= A6XX_RB_2D_BLIT_CNTL_COLOR_FORMAT(fmt);
39 blit_cntl |= A6XX_RB_2D_BLIT_CNTL_IFMT(tu6_rb_fmt_to_ifmt(fmt));
40 return blit_cntl;
41 }
42
43 static void
44 tu_dma_prepare(struct tu_cmd_buffer *cmdbuf)
45 {
46 tu_cs_reserve_space(cmdbuf->device, &cmdbuf->cs, 10);
47
48 tu_cs_emit_pkt7(&cmdbuf->cs, CP_EVENT_WRITE, 1);
49 tu_cs_emit(&cmdbuf->cs, PC_CCU_INVALIDATE_COLOR);
50
51 tu_cs_emit_pkt7(&cmdbuf->cs, CP_EVENT_WRITE, 1);
52 tu_cs_emit(&cmdbuf->cs, LRZ_FLUSH);
53
54 tu_cs_emit_pkt7(&cmdbuf->cs, CP_SKIP_IB2_ENABLE_GLOBAL, 1);
55 tu_cs_emit(&cmdbuf->cs, 0x0);
56
57 tu_cs_emit_wfi(&cmdbuf->cs);
58
59 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_RB_CCU_CNTL, 1);
60 tu_cs_emit(&cmdbuf->cs, 0x10000000);
61 }
62
63 static void
64 tu_copy_buffer(struct tu_cmd_buffer *cmdbuf,
65 struct tu_bo *src_bo,
66 uint64_t src_offset,
67 struct tu_bo *dst_bo,
68 uint64_t dst_offset,
69 uint64_t size)
70 {
71 const unsigned max_size_per_iter = 0x4000 - 0x40;
72 const unsigned max_iterations =
73 (size + max_size_per_iter) / max_size_per_iter;
74
75 tu_bo_list_add(&cmdbuf->bo_list, src_bo, MSM_SUBMIT_BO_READ);
76 tu_bo_list_add(&cmdbuf->bo_list, dst_bo, MSM_SUBMIT_BO_WRITE);
77
78 tu_dma_prepare(cmdbuf);
79
80 tu_cs_reserve_space(cmdbuf->device, &cmdbuf->cs, 21 + 48 * max_iterations);
81
82 /* buffer copy setup */
83 tu_cs_emit_pkt7(&cmdbuf->cs, CP_SET_MARKER, 1);
84 tu_cs_emit(&cmdbuf->cs, A2XX_CP_SET_MARKER_0_MODE(RM6_BLIT2DSCALE));
85
86 const uint32_t blit_cntl = blit_control(RB6_R8_UNORM) | 0x20000000;
87
88 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_RB_2D_BLIT_CNTL, 1);
89 tu_cs_emit(&cmdbuf->cs, blit_cntl);
90
91 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_GRAS_2D_BLIT_CNTL, 1);
92 tu_cs_emit(&cmdbuf->cs, blit_cntl);
93
94 for (; size;) {
95 uint64_t src_va = src_bo->iova + src_offset;
96 uint64_t dst_va = dst_bo->iova + dst_offset;
97
98 unsigned src_shift = src_va & 0x3f;
99 unsigned dst_shift = dst_va & 0x3f;
100 unsigned max_shift = MAX2(src_shift, dst_shift);
101
102 src_va -= src_shift;
103 dst_va -= dst_shift;
104
105 uint32_t size_todo = MIN2(0x4000 - max_shift, size);
106 unsigned pitch = (size_todo + max_shift + 63) & ~63;
107
108 /*
109 * Emit source:
110 */
111 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_SP_PS_2D_SRC_INFO, 13);
112 tu_cs_emit(&cmdbuf->cs,
113 A6XX_SP_PS_2D_SRC_INFO_COLOR_FORMAT(RB6_R8_UNORM) |
114 A6XX_SP_PS_2D_SRC_INFO_TILE_MODE(TILE6_LINEAR) |
115 A6XX_SP_PS_2D_SRC_INFO_COLOR_SWAP(WZYX) | 0x500000);
116 tu_cs_emit(&cmdbuf->cs,
117 A6XX_SP_PS_2D_SRC_SIZE_WIDTH(src_shift + size_todo) |
118 A6XX_SP_PS_2D_SRC_SIZE_HEIGHT(1)); /* SP_PS_2D_SRC_SIZE */
119 tu_cs_emit_qw(&cmdbuf->cs, src_va);
120 tu_cs_emit(&cmdbuf->cs, A6XX_SP_PS_2D_SRC_PITCH_PITCH(pitch));
121
122 tu_cs_emit(&cmdbuf->cs, 0x00000000);
123 tu_cs_emit(&cmdbuf->cs, 0x00000000);
124 tu_cs_emit(&cmdbuf->cs, 0x00000000);
125 tu_cs_emit(&cmdbuf->cs, 0x00000000);
126 tu_cs_emit(&cmdbuf->cs, 0x00000000);
127
128 tu_cs_emit(&cmdbuf->cs, 0x00000000);
129 tu_cs_emit(&cmdbuf->cs, 0x00000000);
130 tu_cs_emit(&cmdbuf->cs, 0x00000000);
131
132 /*
133 * Emit destination:
134 */
135 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_RB_2D_DST_INFO, 9);
136 tu_cs_emit(&cmdbuf->cs, A6XX_RB_2D_DST_INFO_COLOR_FORMAT(RB6_R8_UNORM) |
137 A6XX_RB_2D_DST_INFO_TILE_MODE(TILE6_LINEAR) |
138 A6XX_RB_2D_DST_INFO_COLOR_SWAP(WZYX));
139 tu_cs_emit_qw(&cmdbuf->cs, dst_va);
140
141 tu_cs_emit(&cmdbuf->cs, A6XX_RB_2D_DST_SIZE_PITCH(pitch));
142 tu_cs_emit(&cmdbuf->cs, 0x00000000);
143 tu_cs_emit(&cmdbuf->cs, 0x00000000);
144 tu_cs_emit(&cmdbuf->cs, 0x00000000);
145 tu_cs_emit(&cmdbuf->cs, 0x00000000);
146 tu_cs_emit(&cmdbuf->cs, 0x00000000);
147
148 /*
149 * Blit command:
150 */
151 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_GRAS_2D_SRC_TL_X, 4);
152 tu_cs_emit(&cmdbuf->cs, A6XX_GRAS_2D_SRC_TL_X_X(src_shift));
153 tu_cs_emit(&cmdbuf->cs,
154 A6XX_GRAS_2D_SRC_BR_X_X(src_shift + size_todo - 1));
155 tu_cs_emit(&cmdbuf->cs, A6XX_GRAS_2D_SRC_TL_Y_Y(0));
156 tu_cs_emit(&cmdbuf->cs, A6XX_GRAS_2D_SRC_BR_Y_Y(0));
157
158 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_GRAS_2D_DST_TL, 2);
159 tu_cs_emit(&cmdbuf->cs,
160 A6XX_GRAS_2D_DST_TL_X(dst_shift) | A6XX_GRAS_2D_DST_TL_Y(0));
161 tu_cs_emit(&cmdbuf->cs,
162 A6XX_GRAS_2D_DST_BR_X(dst_shift + size_todo - 1) |
163 A6XX_GRAS_2D_DST_BR_Y(0));
164
165 tu_cs_emit_pkt7(&cmdbuf->cs, CP_EVENT_WRITE, 1);
166 tu_cs_emit(&cmdbuf->cs, 0x3f);
167 tu_cs_emit_wfi(&cmdbuf->cs);
168
169 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_RB_UNKNOWN_8C01, 1);
170 tu_cs_emit(&cmdbuf->cs, 0);
171
172 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_SP_2D_SRC_FORMAT, 1);
173 tu_cs_emit(&cmdbuf->cs, 0xf180);
174
175 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_RB_UNKNOWN_8E04, 1);
176 tu_cs_emit(&cmdbuf->cs, 0x01000000);
177
178 tu_cs_emit_pkt7(&cmdbuf->cs, CP_BLIT, 1);
179 tu_cs_emit(&cmdbuf->cs, CP_BLIT_0_OP(BLIT_OP_SCALE));
180
181 tu_cs_emit_wfi(&cmdbuf->cs);
182
183 tu_cs_emit_pkt4(&cmdbuf->cs, REG_A6XX_RB_UNKNOWN_8E04, 1);
184 tu_cs_emit(&cmdbuf->cs, 0);
185
186 src_offset += size_todo;
187 dst_offset += size_todo;
188 size -= size_todo;
189 }
190
191 tu6_emit_event_write(cmdbuf, &cmdbuf->cs, 0x1d, true);
192 tu6_emit_event_write(cmdbuf, &cmdbuf->cs, FACENESS_FLUSH, true);
193 tu6_emit_event_write(cmdbuf, &cmdbuf->cs, CACHE_FLUSH_TS, true);
194 }
195
196 void
197 tu_CmdCopyBuffer(VkCommandBuffer commandBuffer,
198 VkBuffer srcBuffer,
199 VkBuffer destBuffer,
200 uint32_t regionCount,
201 const VkBufferCopy *pRegions)
202 {
203 TU_FROM_HANDLE(tu_cmd_buffer, cmdbuf, commandBuffer);
204 TU_FROM_HANDLE(tu_buffer, src_buffer, srcBuffer);
205 TU_FROM_HANDLE(tu_buffer, dst_buffer, destBuffer);
206
207 for (unsigned i = 0; i < regionCount; ++i) {
208 uint64_t src_offset = src_buffer->bo_offset + pRegions[i].srcOffset;
209 uint64_t dst_offset = dst_buffer->bo_offset + pRegions[i].dstOffset;
210
211 tu_copy_buffer(cmdbuf, src_buffer->bo, src_offset, dst_buffer->bo,
212 dst_offset, pRegions[i].size);
213 }
214 }
215
216 static void
217 meta_copy_buffer_to_image(struct tu_cmd_buffer *cmd_buffer,
218 struct tu_buffer *buffer,
219 struct tu_image *image,
220 VkImageLayout layout,
221 uint32_t regionCount,
222 const VkBufferImageCopy *pRegions)
223 {
224 }
225
226 void
227 tu_CmdCopyBufferToImage(VkCommandBuffer commandBuffer,
228 VkBuffer srcBuffer,
229 VkImage destImage,
230 VkImageLayout destImageLayout,
231 uint32_t regionCount,
232 const VkBufferImageCopy *pRegions)
233 {
234 TU_FROM_HANDLE(tu_cmd_buffer, cmd_buffer, commandBuffer);
235 TU_FROM_HANDLE(tu_image, dest_image, destImage);
236 TU_FROM_HANDLE(tu_buffer, src_buffer, srcBuffer);
237
238 meta_copy_buffer_to_image(cmd_buffer, src_buffer, dest_image,
239 destImageLayout, regionCount, pRegions);
240 }
241
242 static void
243 meta_copy_image_to_buffer(struct tu_cmd_buffer *cmd_buffer,
244 struct tu_buffer *buffer,
245 struct tu_image *image,
246 VkImageLayout layout,
247 uint32_t regionCount,
248 const VkBufferImageCopy *pRegions)
249 {
250 }
251
252 void
253 tu_CmdCopyImageToBuffer(VkCommandBuffer commandBuffer,
254 VkImage srcImage,
255 VkImageLayout srcImageLayout,
256 VkBuffer destBuffer,
257 uint32_t regionCount,
258 const VkBufferImageCopy *pRegions)
259 {
260 TU_FROM_HANDLE(tu_cmd_buffer, cmd_buffer, commandBuffer);
261 TU_FROM_HANDLE(tu_image, src_image, srcImage);
262 TU_FROM_HANDLE(tu_buffer, dst_buffer, destBuffer);
263
264 meta_copy_image_to_buffer(cmd_buffer, dst_buffer, src_image,
265 srcImageLayout, regionCount, pRegions);
266 }
267
268 static void
269 meta_copy_image(struct tu_cmd_buffer *cmd_buffer,
270 struct tu_image *src_image,
271 VkImageLayout src_image_layout,
272 struct tu_image *dest_image,
273 VkImageLayout dest_image_layout,
274 uint32_t regionCount,
275 const VkImageCopy *pRegions)
276 {
277 }
278
279 void
280 tu_CmdCopyImage(VkCommandBuffer commandBuffer,
281 VkImage srcImage,
282 VkImageLayout srcImageLayout,
283 VkImage destImage,
284 VkImageLayout destImageLayout,
285 uint32_t regionCount,
286 const VkImageCopy *pRegions)
287 {
288 TU_FROM_HANDLE(tu_cmd_buffer, cmd_buffer, commandBuffer);
289 TU_FROM_HANDLE(tu_image, src_image, srcImage);
290 TU_FROM_HANDLE(tu_image, dest_image, destImage);
291
292 meta_copy_image(cmd_buffer, src_image, srcImageLayout, dest_image,
293 destImageLayout, regionCount, pRegions);
294 }