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
28 #include "util/format/u_format.h"
29 #include "vk_enum_to_str.h"
30 #include "vk_enum_defines.h"
31 #include "vk_util.h"
32
33 #define SWIZ(x,y,z,w) { \
34 PIPE_SWIZZLE_##x, \
35 PIPE_SWIZZLE_##y, \
36 PIPE_SWIZZLE_##z, \
37 PIPE_SWIZZLE_##w \
38 }
39
40 #define FORMAT(vk, rt, tex, swiz, return_size, supports_filtering) \
41 [VK_ENUM_OFFSET(VK_FORMAT_##vk)] = { \
42 1, \
43 {{ \
44 V3D_OUTPUT_IMAGE_FORMAT_##rt, \
45 TEXTURE_DATA_FORMAT_##tex, \
46 swiz, \
47 return_size, \
48 }}, \
49 supports_filtering, \
50 }
51
52 #define PLANE(rt, tex, swiz, return_size) \
53 { \
54 V3D_OUTPUT_IMAGE_FORMAT_##rt, \
55 TEXTURE_DATA_FORMAT_##tex, \
56 swiz, \
57 return_size \
58 }
59
60 #define YCBCR_FORMAT(vk, supports_filtering, plane_count, ...) \
61 [VK_ENUM_OFFSET(VK_FORMAT_##vk)] = { \
62 plane_count, \
63 { \
64 __VA_ARGS__, \
65 }, \
66 supports_filtering, \
67 }
68
69 #define SWIZ_X001 SWIZ(X, 0, 0, 1)
70 #define SWIZ_XY01 SWIZ(X, Y, 0, 1)
71 #define SWIZ_XYZ1 SWIZ(X, Y, Z, 1)
72 #define SWIZ_XYZW SWIZ(X, Y, Z, W)
73 #define SWIZ_YZWX SWIZ(Y, Z, W, X)
74 #define SWIZ_YZW1 SWIZ(Y, Z, W, 1)
75 #define SWIZ_ZYXW SWIZ(Z, Y, X, W)
76 #define SWIZ_ZYX1 SWIZ(Z, Y, X, 1)
77 #define SWIZ_XXXY SWIZ(X, X, X, Y)
78 #define SWIZ_XXX1 SWIZ(X, X, X, 1)
79 #define SWIZ_XXXX SWIZ(X, X, X, X)
80 #define SWIZ_000X SWIZ(0, 0, 0, X)
81 #define SWIZ_WXYZ SWIZ(W, X, Y, Z)
82 #define SWIZ_WZYX SWIZ(W, Z, Y, X)
83
84 /* FIXME: expand format table to describe whether the format is supported
85 * for buffer surfaces (texel buffers, vertex buffers, etc).
86 */
87 static const struct v3dv_format format_table[] = {
88 /* Color, 4 channels */
89 FORMAT(B8G8R8A8_SRGB, SRGB8_ALPHA8, RGBA8, SWIZ_ZYXW, 16, true),
90 FORMAT(B8G8R8A8_UNORM, RGBA8, RGBA8, SWIZ_ZYXW, 16, true),
91
92 FORMAT(R8G8B8A8_SRGB, SRGB8_ALPHA8, RGBA8, SWIZ_XYZW, 16, true),
93 FORMAT(R8G8B8A8_UNORM, RGBA8, RGBA8, SWIZ_XYZW, 16, true),
94 FORMAT(R8G8B8A8_SNORM, NO, RGBA8_SNORM, SWIZ_XYZW, 16, true),
95 FORMAT(R8G8B8A8_SINT, RGBA8I, RGBA8I, SWIZ_XYZW, 16, false),
96 FORMAT(R8G8B8A8_UINT, RGBA8UI, RGBA8UI, SWIZ_XYZW, 16, false),
97
98 FORMAT(R16G16B16A16_SFLOAT, RGBA16F, RGBA16F, SWIZ_XYZW, 16, true),
99 FORMAT(R16G16B16A16_UNORM, NO, RGBA16, SWIZ_XYZW, 32, true),
100 FORMAT(R16G16B16A16_SNORM, NO, RGBA16_SNORM, SWIZ_XYZW, 32, true),
101 FORMAT(R16G16B16A16_SINT, RGBA16I, RGBA16I, SWIZ_XYZW, 16, false),
102 FORMAT(R16G16B16A16_UINT, RGBA16UI, RGBA16UI, SWIZ_XYZW, 16, false),
103
104 FORMAT(R32G32B32A32_SFLOAT, RGBA32F, RGBA32F, SWIZ_XYZW, 32, false),
105 FORMAT(R32G32B32A32_SINT, RGBA32I, RGBA32I, SWIZ_XYZW, 32, false),
106 FORMAT(R32G32B32A32_UINT, RGBA32UI, RGBA32UI, SWIZ_XYZW, 32, false),
107
108 /* Color, 3 channels */
109 FORMAT(R32G32B32_SFLOAT, NO, NO, SWIZ_XYZ1, 0, false),
110 FORMAT(R32G32B32_UINT, NO, NO, SWIZ_XYZ1, 0, false),
111 FORMAT(R32G32B32_SINT, NO, NO, SWIZ_XYZ1, 0, false),
112
113 /* Color, 2 channels */
114 FORMAT(R8G8_UNORM, RG8, RG8, SWIZ_XY01, 16, true),
115 FORMAT(R8G8_SNORM, NO, RG8_SNORM, SWIZ_XY01, 16, true),
116 FORMAT(R8G8_SINT, RG8I, RG8I, SWIZ_XY01, 16, false),
117 FORMAT(R8G8_UINT, RG8UI, RG8UI, SWIZ_XY01, 16, false),
118
119 FORMAT(R16G16_UNORM, NO, RG16, SWIZ_XY01, 32, true),
120 FORMAT(R16G16_SNORM, NO, RG16_SNORM, SWIZ_XY01, 32, true),
121 FORMAT(R16G16_SFLOAT, RG16F, RG16F, SWIZ_XY01, 16, true),
122 FORMAT(R16G16_SINT, RG16I, RG16I, SWIZ_XY01, 16, false),
123 FORMAT(R16G16_UINT, RG16UI, RG16UI, SWIZ_XY01, 16, false),
124
125 FORMAT(R32G32_SFLOAT, RG32F, RG32F, SWIZ_XY01, 32, false),
126 FORMAT(R32G32_SINT, RG32I, RG32I, SWIZ_XY01, 32, false),
127 FORMAT(R32G32_UINT, RG32UI, RG32UI, SWIZ_XY01, 32, false),
128
129 /* Color, 1 channel */
130 FORMAT(R8_UNORM, R8, R8, SWIZ_X001, 16, true),
131 FORMAT(R8_SNORM, NO, R8_SNORM, SWIZ_X001, 16, true),
132 FORMAT(R8_SINT, R8I, R8I, SWIZ_X001, 16, false),
133 FORMAT(R8_UINT, R8UI, R8UI, SWIZ_X001, 16, false),
134
135 FORMAT(R16_UNORM, NO, R16, SWIZ_X001, 32, true),
136 FORMAT(R16_SNORM, NO, R16_SNORM, SWIZ_X001, 32, true),
137 FORMAT(R16_SFLOAT, R16F, R16F, SWIZ_X001, 16, true),
138 FORMAT(R16_SINT, R16I, R16I, SWIZ_X001, 16, false),
139 FORMAT(R16_UINT, R16UI, R16UI, SWIZ_X001, 16, false),
140
141 FORMAT(R32_SFLOAT, R32F, R32F, SWIZ_X001, 32, false),
142 FORMAT(R32_SINT, R32I, R32I, SWIZ_X001, 32, false),
143 FORMAT(R32_UINT, R32UI, R32UI, SWIZ_X001, 32, false),
144
145 /* Color, packed */
146 FORMAT(R4G4B4A4_UNORM_PACK16, ABGR4444, RGBA4, SWIZ_XYZW, 16, true),
147 FORMAT(B4G4R4A4_UNORM_PACK16, ABGR4444, RGBA4, SWIZ_ZYXW, 16, true), /* Swap RB */
148 FORMAT(R5G6B5_UNORM_PACK16, BGR565, RGB565, SWIZ_XYZ1, 16, true),
149 FORMAT(R5G5B5A1_UNORM_PACK16, ABGR1555, RGB5_A1, SWIZ_XYZW, 16, true),
150 FORMAT(A1R5G5B5_UNORM_PACK16, RGBA5551, A1_RGB5, SWIZ_ZYXW, 16, true), /* Swap RB */
151 FORMAT(A8B8G8R8_UNORM_PACK32, RGBA8, RGBA8, SWIZ_XYZW, 16, true), /* RGBA8 UNORM */
152 FORMAT(A8B8G8R8_SNORM_PACK32, NO, RGBA8_SNORM, SWIZ_XYZW, 16, true), /* RGBA8 SNORM */
153 FORMAT(A8B8G8R8_UINT_PACK32, RGBA8UI, RGBA8UI, SWIZ_XYZW, 16, false), /* RGBA8 UINT */
154 FORMAT(A8B8G8R8_SINT_PACK32, RGBA8I, RGBA8I, SWIZ_XYZW, 16, false), /* RGBA8 SINT */
155 FORMAT(A8B8G8R8_SRGB_PACK32, SRGB8_ALPHA8, RGBA8, SWIZ_XYZW, 16, true), /* RGBA8 sRGB */
156 FORMAT(A2B10G10R10_UNORM_PACK32,RGB10_A2, RGB10_A2, SWIZ_XYZW, 16, true),
157 FORMAT(A2B10G10R10_UINT_PACK32, RGB10_A2UI, RGB10_A2UI, SWIZ_XYZW, 16, false),
158 FORMAT(A2R10G10B10_UNORM_PACK32,RGB10_A2, RGB10_A2, SWIZ_ZYXW, 16, true),
159 FORMAT(E5B9G9R9_UFLOAT_PACK32, NO, RGB9_E5, SWIZ_XYZ1, 16, true),
160 FORMAT(B10G11R11_UFLOAT_PACK32, R11F_G11F_B10F,R11F_G11F_B10F, SWIZ_XYZ1, 16, true),
161
162 /* Depth */
163 FORMAT(D16_UNORM, D16, DEPTH_COMP16, SWIZ_X001, 32, false),
164 FORMAT(D32_SFLOAT, D32F, DEPTH_COMP32F, SWIZ_X001, 32, false),
165 FORMAT(X8_D24_UNORM_PACK32, D24S8, DEPTH24_X8, SWIZ_X001, 32, false),
166
167 /* Depth + Stencil */
168 FORMAT(D24_UNORM_S8_UINT, D24S8, DEPTH24_X8, SWIZ_X001, 32, false),
169
170 /* Compressed: ETC2 / EAC */
171 FORMAT(ETC2_R8G8B8_UNORM_BLOCK, NO, RGB8_ETC2, SWIZ_XYZ1, 16, true),
172 FORMAT(ETC2_R8G8B8_SRGB_BLOCK, NO, RGB8_ETC2, SWIZ_XYZ1, 16, true),
173 FORMAT(ETC2_R8G8B8A1_UNORM_BLOCK, NO, RGB8_PUNCHTHROUGH_ALPHA1, SWIZ_XYZW, 16, true),
174 FORMAT(ETC2_R8G8B8A1_SRGB_BLOCK, NO, RGB8_PUNCHTHROUGH_ALPHA1, SWIZ_XYZW, 16, true),
175 FORMAT(ETC2_R8G8B8A8_UNORM_BLOCK, NO, RGBA8_ETC2_EAC, SWIZ_XYZW, 16, true),
176 FORMAT(ETC2_R8G8B8A8_SRGB_BLOCK, NO, RGBA8_ETC2_EAC, SWIZ_XYZW, 16, true),
177 FORMAT(EAC_R11_UNORM_BLOCK, NO, R11_EAC, SWIZ_X001, 16, true),
178 FORMAT(EAC_R11_SNORM_BLOCK, NO, SIGNED_R11_EAC, SWIZ_X001, 16, true),
179 FORMAT(EAC_R11G11_UNORM_BLOCK, NO, RG11_EAC, SWIZ_XY01, 16, true),
180 FORMAT(EAC_R11G11_SNORM_BLOCK, NO, SIGNED_RG11_EAC, SWIZ_XY01, 16, true),
181
182 /* Compressed: BC1-3 */
183 FORMAT(BC1_RGB_UNORM_BLOCK, NO, BC1, SWIZ_XYZ1, 16, true),
184 FORMAT(BC1_RGB_SRGB_BLOCK, NO, BC1, SWIZ_XYZ1, 16, true),
185 FORMAT(BC1_RGBA_UNORM_BLOCK, NO, BC1, SWIZ_XYZW, 16, true),
186 FORMAT(BC1_RGBA_SRGB_BLOCK, NO, BC1, SWIZ_XYZW, 16, true),
187 FORMAT(BC2_UNORM_BLOCK, NO, BC2, SWIZ_XYZW, 16, true),
188 FORMAT(BC2_SRGB_BLOCK, NO, BC2, SWIZ_XYZW, 16, true),
189 FORMAT(BC3_UNORM_BLOCK, NO, BC3, SWIZ_XYZW, 16, true),
190 FORMAT(BC3_SRGB_BLOCK, NO, BC3, SWIZ_XYZW, 16, true),
191
192 /* Compressed: ASTC */
193 FORMAT(ASTC_4x4_UNORM_BLOCK, NO, ASTC_4X4, SWIZ_XYZW, 16, true),
194 FORMAT(ASTC_4x4_SRGB_BLOCK, NO, ASTC_4X4, SWIZ_XYZW, 16, true),
195 FORMAT(ASTC_5x4_UNORM_BLOCK, NO, ASTC_5X4, SWIZ_XYZW, 16, true),
196 FORMAT(ASTC_5x4_SRGB_BLOCK, NO, ASTC_5X4, SWIZ_XYZW, 16, true),
197 FORMAT(ASTC_5x5_UNORM_BLOCK, NO, ASTC_5X5, SWIZ_XYZW, 16, true),
198 FORMAT(ASTC_5x5_SRGB_BLOCK, NO, ASTC_5X5, SWIZ_XYZW, 16, true),
199 FORMAT(ASTC_6x5_UNORM_BLOCK, NO, ASTC_6X5, SWIZ_XYZW, 16, true),
200 FORMAT(ASTC_6x5_SRGB_BLOCK, NO, ASTC_6X5, SWIZ_XYZW, 16, true),
201 FORMAT(ASTC_6x6_UNORM_BLOCK, NO, ASTC_6X6, SWIZ_XYZW, 16, true),
202 FORMAT(ASTC_6x6_SRGB_BLOCK, NO, ASTC_6X6, SWIZ_XYZW, 16, true),
203 FORMAT(ASTC_8x5_UNORM_BLOCK, NO, ASTC_8X5, SWIZ_XYZW, 16, true),
204 FORMAT(ASTC_8x5_SRGB_BLOCK, NO, ASTC_8X5, SWIZ_XYZW, 16, true),
205 FORMAT(ASTC_8x6_UNORM_BLOCK, NO, ASTC_8X6, SWIZ_XYZW, 16, true),
206 FORMAT(ASTC_8x6_SRGB_BLOCK, NO, ASTC_8X6, SWIZ_XYZW, 16, true),
207 FORMAT(ASTC_8x8_UNORM_BLOCK, NO, ASTC_8X8, SWIZ_XYZW, 16, true),
208 FORMAT(ASTC_8x8_SRGB_BLOCK, NO, ASTC_8X8, SWIZ_XYZW, 16, true),
209 FORMAT(ASTC_10x5_UNORM_BLOCK, NO, ASTC_10X5, SWIZ_XYZW, 16, true),
210 FORMAT(ASTC_10x5_SRGB_BLOCK, NO, ASTC_10X5, SWIZ_XYZW, 16, true),
211 FORMAT(ASTC_10x6_UNORM_BLOCK, NO, ASTC_10X6, SWIZ_XYZW, 16, true),
212 FORMAT(ASTC_10x6_SRGB_BLOCK, NO, ASTC_10X6, SWIZ_XYZW, 16, true),
213 FORMAT(ASTC_10x8_UNORM_BLOCK, NO, ASTC_10X8, SWIZ_XYZW, 16, true),
214 FORMAT(ASTC_10x8_SRGB_BLOCK, NO, ASTC_10X8, SWIZ_XYZW, 16, true),
215 FORMAT(ASTC_10x10_UNORM_BLOCK, NO, ASTC_10X10, SWIZ_XYZW, 16, true),
216 FORMAT(ASTC_10x10_SRGB_BLOCK, NO, ASTC_10X10, SWIZ_XYZW, 16, true),
217 FORMAT(ASTC_12x10_UNORM_BLOCK, NO, ASTC_12X10, SWIZ_XYZW, 16, true),
218 FORMAT(ASTC_12x10_SRGB_BLOCK, NO, ASTC_12X10, SWIZ_XYZW, 16, true),
219 FORMAT(ASTC_12x12_UNORM_BLOCK, NO, ASTC_12X12, SWIZ_XYZW, 16, true),
220 FORMAT(ASTC_12x12_SRGB_BLOCK, NO, ASTC_12X12, SWIZ_XYZW, 16, true),
221 };
222
223 /**
224 * Vulkan layout for 4444 formats is defined like this:
225 *
226 * Vulkan ABGR4: (LSB) R | G | B | A (MSB)
227 * Vulkan ARGB4: (LSB) B | G | R | A (MSB)
228 *
229 * We map this to the V3D RGB4 texture format, which really, is ABGR4 with
230 * R in the MSB, so:
231 *
232 * V3D ABGR4 : (LSB) A | B | G | R (MSB)
233 *
234 * Which is reversed from Vulkan's ABGR4 layout. So in order to match Vulkan
235 * semantics we need to apply the following swizzles:
236 *
237 * ABGR4: WZYX (reverse)
238 * ARGB4: YZWX (reverse + swap R/B)
239 */
240 static const struct v3dv_format format_table_4444[] = {
241 FORMAT(A4B4G4R4_UNORM_PACK16, ABGR4444, RGBA4, SWIZ_WZYX, 16, true), /* Reverse */
242 FORMAT(A4R4G4B4_UNORM_PACK16, ABGR4444, RGBA4, SWIZ_YZWX, 16, true), /* Reverse + RB swap */
243 };
244
245 /* VK_KHR_maintenance5 introduces A1B5G5R5 and A8 but we only support the
246 * former. It might be possible to support A8 as R8 with special casing
247 * in a number of places but it would probably take some effort.
248 */
249 static const struct v3dv_format format_table_maintenance5[] = {
250 FORMAT(A1B5G5R5_UNORM_PACK16_KHR, RGBA5551, A1_RGB5, SWIZ_XYZW, 16, true),
251 };
252
253 static const struct v3dv_format format_table_ycbcr[] = {
254 YCBCR_FORMAT(G8_B8R8_2PLANE_420_UNORM, false, 2,
255 PLANE(R8, R8, SWIZ(X, 0, 0, 1), 16),
256 PLANE(RG8, RG8, SWIZ(X, Y, 0, 1), 16)
257 ),
258 YCBCR_FORMAT(G8_B8_R8_3PLANE_420_UNORM, false, 3,
259 PLANE(R8, R8, SWIZ(X, 0, 0, 1), 16),
260 PLANE(R8, R8, SWIZ(X, 0, 0, 1), 16),
261 PLANE(R8, R8, SWIZ(X, 0, 0, 1), 16)
262 ),
263 };
264
265 const struct v3dv_format *
v3dX(get_format)266 v3dX(get_format)(VkFormat format)
267 {
268 /* Core formats */
269 if (format < ARRAY_SIZE(format_table) && format_table[format].plane_count)
270 return &format_table[format];
271
272 uint32_t ext_number = VK_ENUM_EXTENSION(format);
273 uint32_t enum_offset = VK_ENUM_OFFSET(format);
274
275 switch (ext_number) {
276 case _VK_EXT_4444_formats_number:
277 if (enum_offset < ARRAY_SIZE(format_table_4444))
278 return &format_table_4444[enum_offset];
279 else
280 return NULL;
281 case _VK_KHR_sampler_ycbcr_conversion_number:
282 if (enum_offset < ARRAY_SIZE(format_table_ycbcr))
283 return &format_table_ycbcr[enum_offset];
284 else
285 return NULL;
286 case _VK_KHR_maintenance5_number:
287 if (enum_offset < ARRAY_SIZE(format_table_maintenance5))
288 return &format_table_maintenance5[enum_offset];
289 else
290 return NULL;
291 default:
292 return NULL;
293 }
294 }
295
296 void
v3dX(get_internal_type_bpp_for_output_format)297 v3dX(get_internal_type_bpp_for_output_format)(uint32_t format,
298 uint32_t *type,
299 uint32_t *bpp)
300 {
301 switch (format) {
302 case V3D_OUTPUT_IMAGE_FORMAT_RGBA8:
303 case V3D_OUTPUT_IMAGE_FORMAT_RGB8:
304 case V3D_OUTPUT_IMAGE_FORMAT_RG8:
305 case V3D_OUTPUT_IMAGE_FORMAT_R8:
306 case V3D_OUTPUT_IMAGE_FORMAT_ABGR4444:
307 case V3D_OUTPUT_IMAGE_FORMAT_BGR565:
308 case V3D_OUTPUT_IMAGE_FORMAT_ABGR1555:
309 *type = V3D_INTERNAL_TYPE_8;
310 *bpp = V3D_INTERNAL_BPP_32;
311 break;
312
313 case V3D_OUTPUT_IMAGE_FORMAT_RGBA8I:
314 case V3D_OUTPUT_IMAGE_FORMAT_RG8I:
315 case V3D_OUTPUT_IMAGE_FORMAT_R8I:
316 *type = V3D_INTERNAL_TYPE_8I;
317 *bpp = V3D_INTERNAL_BPP_32;
318 break;
319
320 case V3D_OUTPUT_IMAGE_FORMAT_RGBA8UI:
321 case V3D_OUTPUT_IMAGE_FORMAT_RG8UI:
322 case V3D_OUTPUT_IMAGE_FORMAT_R8UI:
323 *type = V3D_INTERNAL_TYPE_8UI;
324 *bpp = V3D_INTERNAL_BPP_32;
325 break;
326
327 case V3D_OUTPUT_IMAGE_FORMAT_SRGB8_ALPHA8:
328 case V3D_OUTPUT_IMAGE_FORMAT_SRGB:
329 case V3D_OUTPUT_IMAGE_FORMAT_RGB10_A2:
330 case V3D_OUTPUT_IMAGE_FORMAT_R11F_G11F_B10F:
331 case V3D_OUTPUT_IMAGE_FORMAT_RGBA16F:
332 /* Note that sRGB RTs are stored in the tile buffer at 16F,
333 * and the conversion to sRGB happens at tilebuffer load/store.
334 */
335 *type = V3D_INTERNAL_TYPE_16F;
336 *bpp = V3D_INTERNAL_BPP_64;
337 break;
338
339 case V3D_OUTPUT_IMAGE_FORMAT_RG16F:
340 case V3D_OUTPUT_IMAGE_FORMAT_R16F:
341 *type = V3D_INTERNAL_TYPE_16F;
342 /* Use 64bpp to make sure the TLB doesn't throw away the alpha
343 * channel before alpha test happens.
344 */
345 *bpp = V3D_INTERNAL_BPP_64;
346 break;
347
348 case V3D_OUTPUT_IMAGE_FORMAT_RGBA16I:
349 *type = V3D_INTERNAL_TYPE_16I;
350 *bpp = V3D_INTERNAL_BPP_64;
351 break;
352
353 case V3D_OUTPUT_IMAGE_FORMAT_RG16I:
354 case V3D_OUTPUT_IMAGE_FORMAT_R16I:
355 *type = V3D_INTERNAL_TYPE_16I;
356 *bpp = V3D_INTERNAL_BPP_32;
357 break;
358
359 case V3D_OUTPUT_IMAGE_FORMAT_RGB10_A2UI:
360 case V3D_OUTPUT_IMAGE_FORMAT_RGBA16UI:
361 *type = V3D_INTERNAL_TYPE_16UI;
362 *bpp = V3D_INTERNAL_BPP_64;
363 break;
364
365 case V3D_OUTPUT_IMAGE_FORMAT_RG16UI:
366 case V3D_OUTPUT_IMAGE_FORMAT_R16UI:
367 *type = V3D_INTERNAL_TYPE_16UI;
368 *bpp = V3D_INTERNAL_BPP_32;
369 break;
370
371 case V3D_OUTPUT_IMAGE_FORMAT_RGBA32I:
372 *type = V3D_INTERNAL_TYPE_32I;
373 *bpp = V3D_INTERNAL_BPP_128;
374 break;
375
376 case V3D_OUTPUT_IMAGE_FORMAT_RG32I:
377 *type = V3D_INTERNAL_TYPE_32I;
378 *bpp = V3D_INTERNAL_BPP_64;
379 break;
380
381 case V3D_OUTPUT_IMAGE_FORMAT_R32I:
382 *type = V3D_INTERNAL_TYPE_32I;
383 *bpp = V3D_INTERNAL_BPP_32;
384 break;
385
386 case V3D_OUTPUT_IMAGE_FORMAT_RGBA32UI:
387 *type = V3D_INTERNAL_TYPE_32UI;
388 *bpp = V3D_INTERNAL_BPP_128;
389 break;
390
391 case V3D_OUTPUT_IMAGE_FORMAT_RG32UI:
392 *type = V3D_INTERNAL_TYPE_32UI;
393 *bpp = V3D_INTERNAL_BPP_64;
394 break;
395
396 case V3D_OUTPUT_IMAGE_FORMAT_R32UI:
397 *type = V3D_INTERNAL_TYPE_32UI;
398 *bpp = V3D_INTERNAL_BPP_32;
399 break;
400
401 case V3D_OUTPUT_IMAGE_FORMAT_RGBA32F:
402 *type = V3D_INTERNAL_TYPE_32F;
403 *bpp = V3D_INTERNAL_BPP_128;
404 break;
405
406 case V3D_OUTPUT_IMAGE_FORMAT_RG32F:
407 *type = V3D_INTERNAL_TYPE_32F;
408 *bpp = V3D_INTERNAL_BPP_64;
409 break;
410
411 case V3D_OUTPUT_IMAGE_FORMAT_R32F:
412 *type = V3D_INTERNAL_TYPE_32F;
413 *bpp = V3D_INTERNAL_BPP_32;
414 break;
415
416 default:
417 /* Provide some default values, as we'll be called at RB
418 * creation time, even if an RB with this format isn't supported.
419 */
420 *type = V3D_INTERNAL_TYPE_8;
421 *bpp = V3D_INTERNAL_BPP_32;
422 break;
423 }
424 }
425
426 bool
v3dX(format_supports_tlb_resolve)427 v3dX(format_supports_tlb_resolve)(const struct v3dv_format *format)
428 {
429 uint32_t type, bpp;
430
431 /* Multiplanar images cannot be multisampled:
432 *
433 * "sampleCounts will be set to VK_SAMPLE_COUNT_1_BIT if at least one of
434 * the following conditions is true: (...) format is one of the formats
435 * that require a sampler Y′CBCR conversion (...)"
436 */
437 if (!format->plane_count || format->plane_count > 1)
438 return false;
439
440 v3dX(get_internal_type_bpp_for_output_format)(format->planes[0].rt_type, &type, &bpp);
441 return type == V3D_INTERNAL_TYPE_8 || type == V3D_INTERNAL_TYPE_16F;
442 }
443
444 bool
v3dX(format_supports_blending)445 v3dX(format_supports_blending)(const struct v3dv_format *format)
446 {
447 /* ycbcr formats don't support blending */
448 if (!format->plane_count || format->plane_count > 1)
449 return false;
450
451 /* Hardware blending is only supported on render targets that are configured
452 * 4x8-bit unorm, 2x16-bit float or 4x16-bit float.
453 */
454 uint32_t type, bpp;
455 v3dX(get_internal_type_bpp_for_output_format)(format->planes[0].rt_type, &type, &bpp);
456 switch (type) {
457 case V3D_INTERNAL_TYPE_8:
458 return bpp == V3D_INTERNAL_BPP_32;
459 case V3D_INTERNAL_TYPE_16F:
460 return bpp == V3D_INTERNAL_BPP_32 || V3D_INTERNAL_BPP_64;
461 default:
462 return false;
463 }
464 }
465
466 bool
v3dX(tfu_supports_tex_format)467 v3dX(tfu_supports_tex_format)(uint32_t tex_format)
468 {
469 switch (tex_format) {
470 case TEXTURE_DATA_FORMAT_R8:
471 case TEXTURE_DATA_FORMAT_R8_SNORM:
472 case TEXTURE_DATA_FORMAT_RG8:
473 case TEXTURE_DATA_FORMAT_RG8_SNORM:
474 case TEXTURE_DATA_FORMAT_RGBA8:
475 case TEXTURE_DATA_FORMAT_RGBA8_SNORM:
476 case TEXTURE_DATA_FORMAT_RGB565:
477 case TEXTURE_DATA_FORMAT_RGBA4:
478 case TEXTURE_DATA_FORMAT_RGB5_A1:
479 case TEXTURE_DATA_FORMAT_RGB10_A2:
480 case TEXTURE_DATA_FORMAT_R16:
481 case TEXTURE_DATA_FORMAT_R16_SNORM:
482 case TEXTURE_DATA_FORMAT_RG16:
483 case TEXTURE_DATA_FORMAT_RG16_SNORM:
484 case TEXTURE_DATA_FORMAT_RGBA16:
485 case TEXTURE_DATA_FORMAT_RGBA16_SNORM:
486 case TEXTURE_DATA_FORMAT_R16F:
487 case TEXTURE_DATA_FORMAT_RG16F:
488 case TEXTURE_DATA_FORMAT_RGBA16F:
489 case TEXTURE_DATA_FORMAT_R11F_G11F_B10F:
490 case TEXTURE_DATA_FORMAT_R4:
491 case TEXTURE_DATA_FORMAT_RGB9_E5:
492 case TEXTURE_DATA_FORMAT_R32F:
493 case TEXTURE_DATA_FORMAT_RG32F:
494 case TEXTURE_DATA_FORMAT_RGBA32F:
495 case TEXTURE_DATA_FORMAT_RGB8_ETC2:
496 case TEXTURE_DATA_FORMAT_RGB8_PUNCHTHROUGH_ALPHA1:
497 case TEXTURE_DATA_FORMAT_RGBA8_ETC2_EAC:
498 case TEXTURE_DATA_FORMAT_R11_EAC:
499 case TEXTURE_DATA_FORMAT_SIGNED_R11_EAC:
500 case TEXTURE_DATA_FORMAT_RG11_EAC:
501 case TEXTURE_DATA_FORMAT_SIGNED_RG11_EAC:
502 return true;
503 default:
504 return false;
505 }
506 }
507
508 uint8_t
v3dX(get_internal_depth_type)509 v3dX(get_internal_depth_type)(VkFormat format)
510 {
511 switch (format) {
512 case VK_FORMAT_D16_UNORM:
513 return V3D_INTERNAL_TYPE_DEPTH_16;
514 case VK_FORMAT_D32_SFLOAT:
515 return V3D_INTERNAL_TYPE_DEPTH_32F;
516 case VK_FORMAT_X8_D24_UNORM_PACK32:
517 case VK_FORMAT_D24_UNORM_S8_UINT:
518 return V3D_INTERNAL_TYPE_DEPTH_24;
519 default:
520 unreachable("Invalid depth format");
521 break;
522 }
523 }
524
525 void
v3dX(get_internal_type_bpp_for_image_aspects)526 v3dX(get_internal_type_bpp_for_image_aspects)(VkFormat vk_format,
527 VkImageAspectFlags aspect_mask,
528 uint32_t *internal_type,
529 uint32_t *internal_bpp)
530 {
531 /* We can't store depth/stencil pixel formats to a raster format, so
532 * instead we load our depth/stencil aspects to a compatible color format.
533 */
534 if (aspect_mask & (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
535 *internal_bpp = V3D_INTERNAL_BPP_32;
536 switch (vk_format) {
537 case VK_FORMAT_D16_UNORM:
538 *internal_type = V3D_INTERNAL_TYPE_16UI;
539 break;
540 case VK_FORMAT_D32_SFLOAT:
541 *internal_type = V3D_INTERNAL_TYPE_32F;
542 break;
543 case VK_FORMAT_X8_D24_UNORM_PACK32:
544 case VK_FORMAT_D24_UNORM_S8_UINT:
545 /* Use RGBA8 format so we can relocate the X/S bits in the appropriate
546 * place to match Vulkan expectations. See the comment on the tile
547 * load command for more details.
548 */
549 *internal_type = V3D_INTERNAL_TYPE_8UI;
550 break;
551 default:
552 assert(!"unsupported format");
553 break;
554 }
555 } else {
556 const struct v3dv_format *format = v3dX(get_format)(vk_format);
557 /* We only expect this to be called for single-plane formats */
558 assert(format->plane_count == 1);
559 v3dX(get_internal_type_bpp_for_output_format)(format->planes[0].rt_type,
560 internal_type, internal_bpp);
561 }
562 }
563