1 /*
2 * Copyright © 2016 Red Hat.
3 * Copyright © 2016 Bas Nieuwenhuizen
4 *
5 * based in part on anv driver which is:
6 * Copyright © 2015 Intel Corporation
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the next
16 * paragraph) shall be included in all copies or substantial portions of the
17 * Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
20 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
23 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
24 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
25 * IN THE SOFTWARE.
26 */
27
28 #include "radv_private.h"
29
30 #include "vk_sampler.h"
31
32 static unsigned
radv_tex_wrap(VkSamplerAddressMode address_mode)33 radv_tex_wrap(VkSamplerAddressMode address_mode)
34 {
35 switch (address_mode) {
36 case VK_SAMPLER_ADDRESS_MODE_REPEAT:
37 return V_008F30_SQ_TEX_WRAP;
38 case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT:
39 return V_008F30_SQ_TEX_MIRROR;
40 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE:
41 return V_008F30_SQ_TEX_CLAMP_LAST_TEXEL;
42 case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER:
43 return V_008F30_SQ_TEX_CLAMP_BORDER;
44 case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE:
45 return V_008F30_SQ_TEX_MIRROR_ONCE_LAST_TEXEL;
46 default:
47 unreachable("illegal tex wrap mode");
48 break;
49 }
50 return 0;
51 }
52
53 static unsigned
radv_tex_compare(VkCompareOp op)54 radv_tex_compare(VkCompareOp op)
55 {
56 switch (op) {
57 case VK_COMPARE_OP_NEVER:
58 return V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER;
59 case VK_COMPARE_OP_LESS:
60 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESS;
61 case VK_COMPARE_OP_EQUAL:
62 return V_008F30_SQ_TEX_DEPTH_COMPARE_EQUAL;
63 case VK_COMPARE_OP_LESS_OR_EQUAL:
64 return V_008F30_SQ_TEX_DEPTH_COMPARE_LESSEQUAL;
65 case VK_COMPARE_OP_GREATER:
66 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATER;
67 case VK_COMPARE_OP_NOT_EQUAL:
68 return V_008F30_SQ_TEX_DEPTH_COMPARE_NOTEQUAL;
69 case VK_COMPARE_OP_GREATER_OR_EQUAL:
70 return V_008F30_SQ_TEX_DEPTH_COMPARE_GREATEREQUAL;
71 case VK_COMPARE_OP_ALWAYS:
72 return V_008F30_SQ_TEX_DEPTH_COMPARE_ALWAYS;
73 default:
74 unreachable("illegal compare mode");
75 break;
76 }
77 return 0;
78 }
79
80 static unsigned
radv_tex_filter(VkFilter filter,unsigned max_ansio)81 radv_tex_filter(VkFilter filter, unsigned max_ansio)
82 {
83 switch (filter) {
84 case VK_FILTER_NEAREST:
85 return (max_ansio > 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_POINT : V_008F38_SQ_TEX_XY_FILTER_POINT);
86 case VK_FILTER_LINEAR:
87 return (max_ansio > 1 ? V_008F38_SQ_TEX_XY_FILTER_ANISO_BILINEAR : V_008F38_SQ_TEX_XY_FILTER_BILINEAR);
88 case VK_FILTER_CUBIC_EXT:
89 default:
90 fprintf(stderr, "illegal texture filter");
91 return 0;
92 }
93 }
94
95 static unsigned
radv_tex_mipfilter(VkSamplerMipmapMode mode)96 radv_tex_mipfilter(VkSamplerMipmapMode mode)
97 {
98 switch (mode) {
99 case VK_SAMPLER_MIPMAP_MODE_NEAREST:
100 return V_008F38_SQ_TEX_Z_FILTER_POINT;
101 case VK_SAMPLER_MIPMAP_MODE_LINEAR:
102 return V_008F38_SQ_TEX_Z_FILTER_LINEAR;
103 default:
104 return V_008F38_SQ_TEX_Z_FILTER_NONE;
105 }
106 }
107
108 static unsigned
radv_tex_bordercolor(VkBorderColor bcolor)109 radv_tex_bordercolor(VkBorderColor bcolor)
110 {
111 switch (bcolor) {
112 case VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK:
113 case VK_BORDER_COLOR_INT_TRANSPARENT_BLACK:
114 return V_008F3C_SQ_TEX_BORDER_COLOR_TRANS_BLACK;
115 case VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK:
116 case VK_BORDER_COLOR_INT_OPAQUE_BLACK:
117 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_BLACK;
118 case VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE:
119 case VK_BORDER_COLOR_INT_OPAQUE_WHITE:
120 return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE;
121 case VK_BORDER_COLOR_FLOAT_CUSTOM_EXT:
122 case VK_BORDER_COLOR_INT_CUSTOM_EXT:
123 return V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER;
124 default:
125 break;
126 }
127 return 0;
128 }
129
130 static unsigned
radv_tex_aniso_filter(unsigned filter)131 radv_tex_aniso_filter(unsigned filter)
132 {
133 return MIN2(util_logbase2(filter), 4);
134 }
135
136 static unsigned
radv_tex_filter_mode(VkSamplerReductionMode mode)137 radv_tex_filter_mode(VkSamplerReductionMode mode)
138 {
139 switch (mode) {
140 case VK_SAMPLER_REDUCTION_MODE_WEIGHTED_AVERAGE:
141 return V_008F30_SQ_IMG_FILTER_MODE_BLEND;
142 case VK_SAMPLER_REDUCTION_MODE_MIN:
143 return V_008F30_SQ_IMG_FILTER_MODE_MIN;
144 case VK_SAMPLER_REDUCTION_MODE_MAX:
145 return V_008F30_SQ_IMG_FILTER_MODE_MAX;
146 default:
147 break;
148 }
149 return 0;
150 }
151
152 static uint32_t
radv_get_max_anisotropy(struct radv_device * device,const VkSamplerCreateInfo * pCreateInfo)153 radv_get_max_anisotropy(struct radv_device *device, const VkSamplerCreateInfo *pCreateInfo)
154 {
155 if (device->force_aniso >= 0)
156 return device->force_aniso;
157
158 if (pCreateInfo->anisotropyEnable && pCreateInfo->maxAnisotropy > 1.0f)
159 return (uint32_t)pCreateInfo->maxAnisotropy;
160
161 return 0;
162 }
163
164 static uint32_t
radv_register_border_color(struct radv_device * device,VkClearColorValue value)165 radv_register_border_color(struct radv_device *device, VkClearColorValue value)
166 {
167 uint32_t slot;
168
169 mtx_lock(&device->border_color_data.mutex);
170
171 for (slot = 0; slot < RADV_BORDER_COLOR_COUNT; slot++) {
172 if (!device->border_color_data.used[slot]) {
173 /* Copy to the GPU wrt endian-ness. */
174 util_memcpy_cpu_to_le32(&device->border_color_data.colors_gpu_ptr[slot], &value, sizeof(VkClearColorValue));
175
176 device->border_color_data.used[slot] = true;
177 break;
178 }
179 }
180
181 mtx_unlock(&device->border_color_data.mutex);
182
183 return slot;
184 }
185
186 static void
radv_unregister_border_color(struct radv_device * device,uint32_t slot)187 radv_unregister_border_color(struct radv_device *device, uint32_t slot)
188 {
189 mtx_lock(&device->border_color_data.mutex);
190
191 device->border_color_data.used[slot] = false;
192
193 mtx_unlock(&device->border_color_data.mutex);
194 }
195
196 static void
radv_init_sampler(struct radv_device * device,struct radv_sampler * sampler,const VkSamplerCreateInfo * pCreateInfo)197 radv_init_sampler(struct radv_device *device, struct radv_sampler *sampler, const VkSamplerCreateInfo *pCreateInfo)
198 {
199 uint32_t max_aniso = radv_get_max_anisotropy(device, pCreateInfo);
200 uint32_t max_aniso_ratio = radv_tex_aniso_filter(max_aniso);
201 bool compat_mode =
202 device->physical_device->rad_info.gfx_level == GFX8 || device->physical_device->rad_info.gfx_level == GFX9;
203 unsigned filter_mode = radv_tex_filter_mode(sampler->vk.reduction_mode);
204 unsigned depth_compare_func = V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER;
205 bool trunc_coord = ((pCreateInfo->minFilter == VK_FILTER_NEAREST && pCreateInfo->magFilter == VK_FILTER_NEAREST) ||
206 device->physical_device->rad_info.conformant_trunc_coord) &&
207 !device->disable_trunc_coord;
208 bool uses_border_color = pCreateInfo->addressModeU == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
209 pCreateInfo->addressModeV == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
210 pCreateInfo->addressModeW == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
211 VkBorderColor border_color = uses_border_color ? pCreateInfo->borderColor : VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
212 uint32_t border_color_ptr;
213 bool disable_cube_wrap = pCreateInfo->flags & VK_SAMPLER_CREATE_NON_SEAMLESS_CUBE_MAP_BIT_EXT;
214
215 if (pCreateInfo->compareEnable)
216 depth_compare_func = radv_tex_compare(pCreateInfo->compareOp);
217
218 sampler->border_color_slot = RADV_BORDER_COLOR_COUNT;
219
220 if (vk_border_color_is_custom(border_color)) {
221 sampler->border_color_slot = radv_register_border_color(device, sampler->vk.border_color_value);
222
223 /* Did we fail to find a slot? */
224 if (sampler->border_color_slot == RADV_BORDER_COLOR_COUNT) {
225 fprintf(stderr, "WARNING: no free border color slots, defaulting to TRANS_BLACK.\n");
226 border_color = VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
227 }
228 }
229
230 /* If we don't have a custom color, set the ptr to 0 */
231 border_color_ptr = sampler->border_color_slot != RADV_BORDER_COLOR_COUNT ? sampler->border_color_slot : 0;
232
233 sampler->state[0] = (S_008F30_CLAMP_X(radv_tex_wrap(pCreateInfo->addressModeU)) |
234 S_008F30_CLAMP_Y(radv_tex_wrap(pCreateInfo->addressModeV)) |
235 S_008F30_CLAMP_Z(radv_tex_wrap(pCreateInfo->addressModeW)) |
236 S_008F30_MAX_ANISO_RATIO(max_aniso_ratio) | S_008F30_DEPTH_COMPARE_FUNC(depth_compare_func) |
237 S_008F30_FORCE_UNNORMALIZED(pCreateInfo->unnormalizedCoordinates ? 1 : 0) |
238 S_008F30_ANISO_THRESHOLD(max_aniso_ratio >> 1) | S_008F30_ANISO_BIAS(max_aniso_ratio) |
239 S_008F30_DISABLE_CUBE_WRAP(disable_cube_wrap) | S_008F30_COMPAT_MODE(compat_mode) |
240 S_008F30_FILTER_MODE(filter_mode) | S_008F30_TRUNC_COORD(trunc_coord));
241 sampler->state[1] = (S_008F34_MIN_LOD(radv_float_to_ufixed(CLAMP(pCreateInfo->minLod, 0, 15), 8)) |
242 S_008F34_MAX_LOD(radv_float_to_ufixed(CLAMP(pCreateInfo->maxLod, 0, 15), 8)) |
243 S_008F34_PERF_MIP(max_aniso_ratio ? max_aniso_ratio + 6 : 0));
244 sampler->state[2] = (S_008F38_XY_MAG_FILTER(radv_tex_filter(pCreateInfo->magFilter, max_aniso)) |
245 S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo->minFilter, max_aniso)) |
246 S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo->mipmapMode)));
247 sampler->state[3] = S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(border_color));
248
249 if (device->physical_device->rad_info.gfx_level >= GFX10) {
250 sampler->state[2] |= S_008F38_LOD_BIAS(radv_float_to_sfixed(CLAMP(pCreateInfo->mipLodBias, -32, 31), 8)) |
251 S_008F38_ANISO_OVERRIDE_GFX10(device->instance->drirc.disable_aniso_single_level);
252 } else {
253 sampler->state[2] |= S_008F38_LOD_BIAS(radv_float_to_sfixed(CLAMP(pCreateInfo->mipLodBias, -16, 16), 8)) |
254 S_008F38_DISABLE_LSB_CEIL(device->physical_device->rad_info.gfx_level <= GFX8) |
255 S_008F38_FILTER_PREC_FIX(1) |
256 S_008F38_ANISO_OVERRIDE_GFX8(device->instance->drirc.disable_aniso_single_level &&
257 device->physical_device->rad_info.gfx_level >= GFX8);
258 }
259
260 if (device->physical_device->rad_info.gfx_level >= GFX11) {
261 sampler->state[3] |= S_008F3C_BORDER_COLOR_PTR_GFX11(border_color_ptr);
262 } else {
263 sampler->state[3] |= S_008F3C_BORDER_COLOR_PTR_GFX6(border_color_ptr);
264 }
265 }
266
267 VKAPI_ATTR VkResult VKAPI_CALL
radv_CreateSampler(VkDevice _device,const VkSamplerCreateInfo * pCreateInfo,const VkAllocationCallbacks * pAllocator,VkSampler * pSampler)268 radv_CreateSampler(VkDevice _device, const VkSamplerCreateInfo *pCreateInfo, const VkAllocationCallbacks *pAllocator,
269 VkSampler *pSampler)
270 {
271 RADV_FROM_HANDLE(radv_device, device, _device);
272 struct radv_sampler *sampler;
273
274 sampler = vk_sampler_create(&device->vk, pCreateInfo, pAllocator, sizeof(*sampler));
275 if (!sampler)
276 return vk_error(device, VK_ERROR_OUT_OF_HOST_MEMORY);
277
278 radv_init_sampler(device, sampler, pCreateInfo);
279
280 *pSampler = radv_sampler_to_handle(sampler);
281
282 return VK_SUCCESS;
283 }
284
285 VKAPI_ATTR void VKAPI_CALL
radv_DestroySampler(VkDevice _device,VkSampler _sampler,const VkAllocationCallbacks * pAllocator)286 radv_DestroySampler(VkDevice _device, VkSampler _sampler, const VkAllocationCallbacks *pAllocator)
287 {
288 RADV_FROM_HANDLE(radv_device, device, _device);
289 RADV_FROM_HANDLE(radv_sampler, sampler, _sampler);
290
291 if (!sampler)
292 return;
293
294 if (sampler->border_color_slot != RADV_BORDER_COLOR_COUNT)
295 radv_unregister_border_color(device, sampler->border_color_slot);
296
297 vk_sampler_destroy(&device->vk, pAllocator, &sampler->vk);
298 }
299