1/*
2 * Copyright © 2021 Raspberry Pi
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#include "v3dv_private.h"
25#include "broadcom/common/v3d_macros.h"
26#include "broadcom/cle/v3dx_pack.h"
27#include "broadcom/compiler/v3d_compiler.h"
28
29#include "vk_format_info.h"
30
31/*
32 * This method translates pipe_swizzle to the swizzle values used at the
33 * packet TEXTURE_SHADER_STATE
34 *
35 * FIXME: C&P from v3d, common place?
36 */
37static uint32_t
38translate_swizzle(unsigned char pipe_swizzle)
39{
40   switch (pipe_swizzle) {
41   case PIPE_SWIZZLE_0:
42      return 0;
43   case PIPE_SWIZZLE_1:
44      return 1;
45   case PIPE_SWIZZLE_X:
46   case PIPE_SWIZZLE_Y:
47   case PIPE_SWIZZLE_Z:
48   case PIPE_SWIZZLE_W:
49      return 2 + pipe_swizzle;
50   default:
51      unreachable("unknown swizzle");
52   }
53}
54
55/*
56 * Packs and ensure bo for the shader state (the latter can be temporal).
57 */
58static void
59pack_texture_shader_state_helper(struct v3dv_device *device,
60                                 struct v3dv_image_view *image_view,
61                                 bool for_cube_map_array_storage)
62{
63   assert(!for_cube_map_array_storage ||
64          image_view->vk.view_type == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY);
65   const uint32_t index = for_cube_map_array_storage ? 1 : 0;
66
67   assert(image_view->vk.image);
68   const struct v3dv_image *image = (struct v3dv_image *) image_view->vk.image;
69
70   assert(image->vk.samples == VK_SAMPLE_COUNT_1_BIT ||
71          image->vk.samples == VK_SAMPLE_COUNT_4_BIT);
72   const uint32_t msaa_scale = image->vk.samples == VK_SAMPLE_COUNT_1_BIT ? 1 : 2;
73
74   v3dvx_pack(image_view->texture_shader_state[index], TEXTURE_SHADER_STATE, tex) {
75
76      tex.level_0_is_strictly_uif =
77         (image->slices[0].tiling == V3D_TILING_UIF_XOR ||
78          image->slices[0].tiling == V3D_TILING_UIF_NO_XOR);
79
80      tex.level_0_xor_enable = (image->slices[0].tiling == V3D_TILING_UIF_XOR);
81
82      if (tex.level_0_is_strictly_uif)
83         tex.level_0_ub_pad = image->slices[0].ub_pad;
84
85      /* FIXME: v3d never sets uif_xor_disable, but uses it on the following
86       * check so let's set the default value
87       */
88      tex.uif_xor_disable = false;
89      if (tex.uif_xor_disable ||
90          tex.level_0_is_strictly_uif) {
91         tex.extended = true;
92      }
93
94      tex.base_level = image_view->vk.base_mip_level;
95      tex.max_level = image_view->vk.base_mip_level +
96                      image_view->vk.level_count - 1;
97
98      tex.swizzle_r = translate_swizzle(image_view->swizzle[0]);
99      tex.swizzle_g = translate_swizzle(image_view->swizzle[1]);
100      tex.swizzle_b = translate_swizzle(image_view->swizzle[2]);
101      tex.swizzle_a = translate_swizzle(image_view->swizzle[3]);
102
103      tex.texture_type = image_view->format->tex_type;
104
105      if (image->vk.image_type == VK_IMAGE_TYPE_3D) {
106         tex.image_depth = image->vk.extent.depth;
107      } else {
108         tex.image_depth = image_view->vk.layer_count;
109      }
110
111      /* Empirical testing with CTS shows that when we are sampling from cube
112       * arrays we want to set image depth to layers / 6, but not when doing
113       * image load/store.
114       */
115      if (image_view->vk.view_type == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY &&
116          !for_cube_map_array_storage) {
117         assert(tex.image_depth % 6 == 0);
118         tex.image_depth /= 6;
119      }
120
121      tex.image_height = image->vk.extent.height * msaa_scale;
122      tex.image_width = image->vk.extent.width * msaa_scale;
123
124      /* On 4.x, the height of a 1D texture is redefined to be the
125       * upper 14 bits of the width (which is only usable with txf).
126       */
127      if (image->vk.image_type == VK_IMAGE_TYPE_1D) {
128         tex.image_height = tex.image_width >> 14;
129      }
130      tex.image_width &= (1 << 14) - 1;
131      tex.image_height &= (1 << 14) - 1;
132
133      tex.array_stride_64_byte_aligned = image->cube_map_stride / 64;
134
135      tex.srgb = vk_format_is_srgb(image_view->vk.format);
136
137      /* At this point we don't have the job. That's the reason the first
138       * parameter is NULL, to avoid a crash when cl_pack_emit_reloc tries to
139       * add the bo to the job. This also means that we need to add manually
140       * the image bo to the job using the texture.
141       */
142      const uint32_t base_offset =
143         image->mem->bo->offset +
144         v3dv_layer_offset(image, 0, image_view->vk.base_array_layer);
145      tex.texture_base_pointer = v3dv_cl_address(NULL, base_offset);
146   }
147}
148
149void
150v3dX(pack_texture_shader_state)(struct v3dv_device *device,
151                                struct v3dv_image_view *iview)
152{
153   pack_texture_shader_state_helper(device, iview, false);
154   if (iview->vk.view_type == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY)
155      pack_texture_shader_state_helper(device, iview, true);
156}
157
158void
159v3dX(pack_texture_shader_state_from_buffer_view)(struct v3dv_device *device,
160                                                 struct v3dv_buffer_view *buffer_view)
161{
162   assert(buffer_view->buffer);
163   const struct v3dv_buffer *buffer = buffer_view->buffer;
164
165   v3dvx_pack(buffer_view->texture_shader_state, TEXTURE_SHADER_STATE, tex) {
166      tex.swizzle_r = translate_swizzle(PIPE_SWIZZLE_X);
167      tex.swizzle_g = translate_swizzle(PIPE_SWIZZLE_Y);
168      tex.swizzle_b = translate_swizzle(PIPE_SWIZZLE_Z);
169      tex.swizzle_a = translate_swizzle(PIPE_SWIZZLE_W);
170
171      tex.image_depth = 1;
172
173      /* On 4.x, the height of a 1D texture is redefined to be the upper 14
174       * bits of the width (which is only usable with txf) (or in other words,
175       * we are providing a 28 bit field for size, but split on the usual
176       * 14bit height/width).
177       */
178      tex.image_width = buffer_view->num_elements;
179      tex.image_height = tex.image_width >> 14;
180      tex.image_width &= (1 << 14) - 1;
181      tex.image_height &= (1 << 14) - 1;
182
183      tex.texture_type = buffer_view->format->tex_type;
184      tex.srgb = vk_format_is_srgb(buffer_view->vk_format);
185
186      /* At this point we don't have the job. That's the reason the first
187       * parameter is NULL, to avoid a crash when cl_pack_emit_reloc tries to
188       * add the bo to the job. This also means that we need to add manually
189       * the image bo to the job using the texture.
190       */
191      const uint32_t base_offset =
192         buffer->mem->bo->offset +
193         buffer->mem_offset +
194         buffer_view->offset;
195
196      tex.texture_base_pointer = v3dv_cl_address(NULL, base_offset);
197   }
198}
199