• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright © 2021 Raspberry Pi Ltd
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 /*
30  * Packs and ensure bo for the shader state (the latter can be temporal).
31  */
32 static void
pack_texture_shader_state_helper(struct v3dv_device * device,struct v3dv_image_view * image_view,bool for_cube_map_array_storage)33 pack_texture_shader_state_helper(struct v3dv_device *device,
34                                  struct v3dv_image_view *image_view,
35                                  bool for_cube_map_array_storage)
36 {
37    assert(!for_cube_map_array_storage ||
38           image_view->vk.view_type == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY);
39    const uint32_t index = for_cube_map_array_storage ? 1 : 0;
40 
41    assert(image_view->vk.image);
42    const struct v3dv_image *image = (struct v3dv_image *) image_view->vk.image;
43 
44    assert(image->vk.samples == VK_SAMPLE_COUNT_1_BIT ||
45           image->vk.samples == VK_SAMPLE_COUNT_4_BIT);
46    const uint32_t msaa_scale = image->vk.samples == VK_SAMPLE_COUNT_1_BIT ? 1 : 2;
47 
48    v3dvx_pack(image_view->texture_shader_state[index], TEXTURE_SHADER_STATE, tex) {
49 
50       tex.level_0_is_strictly_uif =
51          (image->slices[0].tiling == V3D_TILING_UIF_XOR ||
52           image->slices[0].tiling == V3D_TILING_UIF_NO_XOR);
53 
54       tex.level_0_xor_enable = (image->slices[0].tiling == V3D_TILING_UIF_XOR);
55 
56       if (tex.level_0_is_strictly_uif)
57          tex.level_0_ub_pad = image->slices[0].ub_pad;
58 
59       /* FIXME: v3d never sets uif_xor_disable, but uses it on the following
60        * check so let's set the default value
61        */
62       tex.uif_xor_disable = false;
63       if (tex.uif_xor_disable ||
64           tex.level_0_is_strictly_uif) {
65          tex.extended = true;
66       }
67 
68       tex.base_level = image_view->vk.base_mip_level;
69       tex.max_level = image_view->vk.base_mip_level +
70                       image_view->vk.level_count - 1;
71 
72       tex.swizzle_r = v3d_translate_pipe_swizzle(image_view->swizzle[0]);
73       tex.swizzle_g = v3d_translate_pipe_swizzle(image_view->swizzle[1]);
74       tex.swizzle_b = v3d_translate_pipe_swizzle(image_view->swizzle[2]);
75       tex.swizzle_a = v3d_translate_pipe_swizzle(image_view->swizzle[3]);
76 
77       tex.reverse_standard_border_color = image_view->channel_reverse;
78 
79       tex.texture_type = image_view->format->tex_type;
80 
81       if (image->vk.image_type == VK_IMAGE_TYPE_3D) {
82          tex.image_depth = image->vk.extent.depth;
83       } else {
84          tex.image_depth = image_view->vk.layer_count;
85       }
86 
87       /* Empirical testing with CTS shows that when we are sampling from cube
88        * arrays we want to set image depth to layers / 6, but not when doing
89        * image load/store.
90        */
91       if (image_view->vk.view_type == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY &&
92           !for_cube_map_array_storage) {
93          assert(tex.image_depth % 6 == 0);
94          tex.image_depth /= 6;
95       }
96 
97       tex.image_height = image->vk.extent.height * msaa_scale;
98       tex.image_width = image->vk.extent.width * msaa_scale;
99 
100       /* On 4.x, the height of a 1D texture is redefined to be the
101        * upper 14 bits of the width (which is only usable with txf).
102        */
103       if (image->vk.image_type == VK_IMAGE_TYPE_1D) {
104          tex.image_height = tex.image_width >> 14;
105       }
106       tex.image_width &= (1 << 14) - 1;
107       tex.image_height &= (1 << 14) - 1;
108 
109       tex.array_stride_64_byte_aligned = image->cube_map_stride / 64;
110 
111       tex.srgb = vk_format_is_srgb(image_view->vk.view_format);
112 
113       /* At this point we don't have the job. That's the reason the first
114        * parameter is NULL, to avoid a crash when cl_pack_emit_reloc tries to
115        * add the bo to the job. This also means that we need to add manually
116        * the image bo to the job using the texture.
117        */
118       const uint32_t base_offset =
119          image->mem->bo->offset +
120          v3dv_layer_offset(image, 0, image_view->vk.base_array_layer);
121       tex.texture_base_pointer = v3dv_cl_address(NULL, base_offset);
122    }
123 }
124 
125 void
v3dX(pack_texture_shader_state)126 v3dX(pack_texture_shader_state)(struct v3dv_device *device,
127                                 struct v3dv_image_view *iview)
128 {
129    pack_texture_shader_state_helper(device, iview, false);
130    if (iview->vk.view_type == VK_IMAGE_VIEW_TYPE_CUBE_ARRAY)
131       pack_texture_shader_state_helper(device, iview, true);
132 }
133 
134 void
v3dX(pack_texture_shader_state_from_buffer_view)135 v3dX(pack_texture_shader_state_from_buffer_view)(struct v3dv_device *device,
136                                                  struct v3dv_buffer_view *buffer_view)
137 {
138    assert(buffer_view->buffer);
139    const struct v3dv_buffer *buffer = buffer_view->buffer;
140 
141    v3dvx_pack(buffer_view->texture_shader_state, TEXTURE_SHADER_STATE, tex) {
142       tex.swizzle_r = v3d_translate_pipe_swizzle(PIPE_SWIZZLE_X);
143       tex.swizzle_g = v3d_translate_pipe_swizzle(PIPE_SWIZZLE_Y);
144       tex.swizzle_b = v3d_translate_pipe_swizzle(PIPE_SWIZZLE_Z);
145       tex.swizzle_a = v3d_translate_pipe_swizzle(PIPE_SWIZZLE_W);
146 
147       tex.image_depth = 1;
148 
149       /* On 4.x, the height of a 1D texture is redefined to be the upper 14
150        * bits of the width (which is only usable with txf) (or in other words,
151        * we are providing a 28 bit field for size, but split on the usual
152        * 14bit height/width).
153        */
154       tex.image_width = buffer_view->num_elements;
155       tex.image_height = tex.image_width >> 14;
156       tex.image_width &= (1 << 14) - 1;
157       tex.image_height &= (1 << 14) - 1;
158 
159       tex.texture_type = buffer_view->format->tex_type;
160       tex.srgb = vk_format_is_srgb(buffer_view->vk_format);
161 
162       /* At this point we don't have the job. That's the reason the first
163        * parameter is NULL, to avoid a crash when cl_pack_emit_reloc tries to
164        * add the bo to the job. This also means that we need to add manually
165        * the image bo to the job using the texture.
166        */
167       const uint32_t base_offset =
168          buffer->mem->bo->offset +
169          buffer->mem_offset +
170          buffer_view->offset;
171 
172       tex.texture_base_pointer = v3dv_cl_address(NULL, base_offset);
173    }
174 }
175