1 /*
2 * Copyright 2020 Valve Corporation
3 * SPDX-License-Identifier: MIT
4 *
5 * Authors:
6 * Jonathan Marek <jonathan@marek.ca>
7 */
8
9 #ifndef TU_UTIL_H
10 #define TU_UTIL_H
11
12 #include <assert.h>
13 #include <stdint.h>
14
15 #include "util/macros.h"
16 #include "util/u_math.h"
17 #include "util/format/u_format_pack.h"
18 #include "util/format/u_format_zs.h"
19 #include "compiler/shader_enums.h"
20
21 #include "adreno_common.xml.h"
22 #include "adreno_pm4.xml.h"
23 #include "a6xx.xml.h"
24
25 #include <vulkan/vulkan.h>
26 #include "vk_util.h"
27
28 #define TU_STAGE_MASK ((1 << MESA_SHADER_STAGES) - 1)
29
30 #define tu_foreach_stage(stage, stage_bits) \
31 for (gl_shader_stage stage, \
32 __tmp = (gl_shader_stage)((stage_bits) &TU_STAGE_MASK); \
33 stage = __builtin_ffs(__tmp) - 1, __tmp; __tmp &= ~(1 << (stage)))
34
35 static inline enum a3xx_msaa_samples
tu_msaa_samples(uint32_t samples)36 tu_msaa_samples(uint32_t samples)
37 {
38 assert(__builtin_popcount(samples) == 1);
39 return util_logbase2(samples);
40 }
41
42 static inline uint32_t
tu6_stage2opcode(gl_shader_stage stage)43 tu6_stage2opcode(gl_shader_stage stage)
44 {
45 if (stage == MESA_SHADER_FRAGMENT || stage == MESA_SHADER_COMPUTE)
46 return CP_LOAD_STATE6_FRAG;
47 return CP_LOAD_STATE6_GEOM;
48 }
49
50 static inline enum a6xx_state_block
tu6_stage2texsb(gl_shader_stage stage)51 tu6_stage2texsb(gl_shader_stage stage)
52 {
53 return SB6_VS_TEX + stage;
54 }
55
56 static inline enum a6xx_state_block
tu6_stage2shadersb(gl_shader_stage stage)57 tu6_stage2shadersb(gl_shader_stage stage)
58 {
59 return SB6_VS_SHADER + stage;
60 }
61
62 static inline enum a3xx_rop_code
tu6_rop(VkLogicOp op)63 tu6_rop(VkLogicOp op)
64 {
65 /* note: hw enum matches the VK enum, but with the 4 bits reversed */
66 static const uint8_t lookup[] = {
67 [VK_LOGIC_OP_CLEAR] = ROP_CLEAR,
68 [VK_LOGIC_OP_AND] = ROP_AND,
69 [VK_LOGIC_OP_AND_REVERSE] = ROP_AND_REVERSE,
70 [VK_LOGIC_OP_COPY] = ROP_COPY,
71 [VK_LOGIC_OP_AND_INVERTED] = ROP_AND_INVERTED,
72 [VK_LOGIC_OP_NO_OP] = ROP_NOOP,
73 [VK_LOGIC_OP_XOR] = ROP_XOR,
74 [VK_LOGIC_OP_OR] = ROP_OR,
75 [VK_LOGIC_OP_NOR] = ROP_NOR,
76 [VK_LOGIC_OP_EQUIVALENT] = ROP_EQUIV,
77 [VK_LOGIC_OP_INVERT] = ROP_INVERT,
78 [VK_LOGIC_OP_OR_REVERSE] = ROP_OR_REVERSE,
79 [VK_LOGIC_OP_COPY_INVERTED] = ROP_COPY_INVERTED,
80 [VK_LOGIC_OP_OR_INVERTED] = ROP_OR_INVERTED,
81 [VK_LOGIC_OP_NAND] = ROP_NAND,
82 [VK_LOGIC_OP_SET] = ROP_SET,
83 };
84 assert(op < ARRAY_SIZE(lookup));
85 return lookup[op];
86 }
87
88 static inline bool
tu6_primtype_line(enum pc_di_primtype type)89 tu6_primtype_line(enum pc_di_primtype type)
90 {
91 switch(type) {
92 case DI_PT_LINELIST:
93 case DI_PT_LINESTRIP:
94 case DI_PT_LINE_ADJ:
95 case DI_PT_LINESTRIP_ADJ:
96 return true;
97 default:
98 return false;
99 }
100 }
101
102 static inline enum pc_di_primtype
tu6_primtype(VkPrimitiveTopology topology)103 tu6_primtype(VkPrimitiveTopology topology)
104 {
105 static const uint8_t lookup[] = {
106 [VK_PRIMITIVE_TOPOLOGY_POINT_LIST] = DI_PT_POINTLIST,
107 [VK_PRIMITIVE_TOPOLOGY_LINE_LIST] = DI_PT_LINELIST,
108 [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP] = DI_PT_LINESTRIP,
109 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST] = DI_PT_TRILIST,
110 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP] = DI_PT_TRISTRIP,
111 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN] = DI_PT_TRIFAN,
112 [VK_PRIMITIVE_TOPOLOGY_LINE_LIST_WITH_ADJACENCY] = DI_PT_LINE_ADJ,
113 [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_WITH_ADJACENCY] = DI_PT_LINESTRIP_ADJ,
114 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_WITH_ADJACENCY] = DI_PT_TRI_ADJ,
115 [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_WITH_ADJACENCY] = DI_PT_TRISTRIP_ADJ,
116 /* Return PATCH0 and update in tu_pipeline_builder_parse_tessellation */
117 [VK_PRIMITIVE_TOPOLOGY_PATCH_LIST] = DI_PT_PATCHES0,
118 };
119 assert(topology < ARRAY_SIZE(lookup));
120 return lookup[topology];
121 }
122
123 static inline enum adreno_compare_func
tu6_compare_func(VkCompareOp op)124 tu6_compare_func(VkCompareOp op)
125 {
126 return (enum adreno_compare_func) op;
127 }
128
129 static inline enum adreno_stencil_op
tu6_stencil_op(VkStencilOp op)130 tu6_stencil_op(VkStencilOp op)
131 {
132 return (enum adreno_stencil_op) op;
133 }
134
135 static inline enum adreno_rb_blend_factor
tu6_blend_factor(VkBlendFactor factor)136 tu6_blend_factor(VkBlendFactor factor)
137 {
138 static const uint8_t lookup[] = {
139 [VK_BLEND_FACTOR_ZERO] = FACTOR_ZERO,
140 [VK_BLEND_FACTOR_ONE] = FACTOR_ONE,
141 [VK_BLEND_FACTOR_SRC_COLOR] = FACTOR_SRC_COLOR,
142 [VK_BLEND_FACTOR_ONE_MINUS_SRC_COLOR] = FACTOR_ONE_MINUS_SRC_COLOR,
143 [VK_BLEND_FACTOR_DST_COLOR] = FACTOR_DST_COLOR,
144 [VK_BLEND_FACTOR_ONE_MINUS_DST_COLOR] = FACTOR_ONE_MINUS_DST_COLOR,
145 [VK_BLEND_FACTOR_SRC_ALPHA] = FACTOR_SRC_ALPHA,
146 [VK_BLEND_FACTOR_ONE_MINUS_SRC_ALPHA] = FACTOR_ONE_MINUS_SRC_ALPHA,
147 [VK_BLEND_FACTOR_DST_ALPHA] = FACTOR_DST_ALPHA,
148 [VK_BLEND_FACTOR_ONE_MINUS_DST_ALPHA] = FACTOR_ONE_MINUS_DST_ALPHA,
149 [VK_BLEND_FACTOR_CONSTANT_COLOR] = FACTOR_CONSTANT_COLOR,
150 [VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_COLOR]= FACTOR_ONE_MINUS_CONSTANT_COLOR,
151 [VK_BLEND_FACTOR_CONSTANT_ALPHA] = FACTOR_CONSTANT_ALPHA,
152 [VK_BLEND_FACTOR_ONE_MINUS_CONSTANT_ALPHA]= FACTOR_ONE_MINUS_CONSTANT_ALPHA,
153 [VK_BLEND_FACTOR_SRC_ALPHA_SATURATE] = FACTOR_SRC_ALPHA_SATURATE,
154 [VK_BLEND_FACTOR_SRC1_COLOR] = FACTOR_SRC1_COLOR,
155 [VK_BLEND_FACTOR_ONE_MINUS_SRC1_COLOR] = FACTOR_ONE_MINUS_SRC1_COLOR,
156 [VK_BLEND_FACTOR_SRC1_ALPHA] = FACTOR_SRC1_ALPHA,
157 [VK_BLEND_FACTOR_ONE_MINUS_SRC1_ALPHA] = FACTOR_ONE_MINUS_SRC1_ALPHA,
158 };
159 assert(factor < ARRAY_SIZE(lookup));
160 return lookup[factor];
161 }
162
163 static inline enum a3xx_rb_blend_opcode
tu6_blend_op(VkBlendOp op)164 tu6_blend_op(VkBlendOp op)
165 {
166 return (enum a3xx_rb_blend_opcode) op;
167 }
168
169 static inline enum a6xx_tex_type
tu6_tex_type(VkImageViewType type,bool storage)170 tu6_tex_type(VkImageViewType type, bool storage)
171 {
172 switch (type) {
173 default:
174 case VK_IMAGE_VIEW_TYPE_1D:
175 case VK_IMAGE_VIEW_TYPE_1D_ARRAY:
176 return A6XX_TEX_1D;
177 case VK_IMAGE_VIEW_TYPE_2D:
178 case VK_IMAGE_VIEW_TYPE_2D_ARRAY:
179 return A6XX_TEX_2D;
180 case VK_IMAGE_VIEW_TYPE_3D:
181 return A6XX_TEX_3D;
182 case VK_IMAGE_VIEW_TYPE_CUBE:
183 case VK_IMAGE_VIEW_TYPE_CUBE_ARRAY:
184 return storage ? A6XX_TEX_2D : A6XX_TEX_CUBE;
185 }
186 }
187
188 static inline enum a6xx_tex_clamp
tu6_tex_wrap(VkSamplerAddressMode address_mode)189 tu6_tex_wrap(VkSamplerAddressMode address_mode)
190 {
191 uint8_t lookup[] = {
192 [VK_SAMPLER_ADDRESS_MODE_REPEAT] = A6XX_TEX_REPEAT,
193 [VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT] = A6XX_TEX_MIRROR_REPEAT,
194 [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE] = A6XX_TEX_CLAMP_TO_EDGE,
195 [VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER] = A6XX_TEX_CLAMP_TO_BORDER,
196 [VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE] = A6XX_TEX_MIRROR_CLAMP,
197 };
198 assert(address_mode < ARRAY_SIZE(lookup));
199 return lookup[address_mode];
200 }
201
202 static inline enum a6xx_tex_filter
tu6_tex_filter(VkFilter filter,unsigned aniso)203 tu6_tex_filter(VkFilter filter, unsigned aniso)
204 {
205 switch (filter) {
206 case VK_FILTER_NEAREST:
207 return A6XX_TEX_NEAREST;
208 case VK_FILTER_LINEAR:
209 return aniso ? A6XX_TEX_ANISO : A6XX_TEX_LINEAR;
210 case VK_FILTER_CUBIC_EXT:
211 return A6XX_TEX_CUBIC;
212 default:
213 unreachable("illegal texture filter");
214 break;
215 }
216 }
217
218 static inline enum a6xx_reduction_mode
tu6_reduction_mode(VkSamplerReductionMode reduction_mode)219 tu6_reduction_mode(VkSamplerReductionMode reduction_mode)
220 {
221 return (enum a6xx_reduction_mode) reduction_mode;
222 }
223
224 static inline enum a6xx_depth_format
tu6_pipe2depth(VkFormat format)225 tu6_pipe2depth(VkFormat format)
226 {
227 switch (format) {
228 case VK_FORMAT_D16_UNORM:
229 return DEPTH6_16;
230 case VK_FORMAT_X8_D24_UNORM_PACK32:
231 case VK_FORMAT_D24_UNORM_S8_UINT:
232 return DEPTH6_24_8;
233 case VK_FORMAT_D32_SFLOAT:
234 case VK_FORMAT_D32_SFLOAT_S8_UINT:
235 case VK_FORMAT_S8_UINT:
236 return DEPTH6_32;
237 default:
238 return ~0;
239 }
240 }
241
242 static inline enum a6xx_polygon_mode
tu6_polygon_mode(VkPolygonMode mode)243 tu6_polygon_mode(VkPolygonMode mode)
244 {
245 switch (mode) {
246 case VK_POLYGON_MODE_POINT:
247 return POLYMODE6_POINTS;
248 case VK_POLYGON_MODE_LINE:
249 return POLYMODE6_LINES;
250 case VK_POLYGON_MODE_FILL:
251 return POLYMODE6_TRIANGLES;
252 default:
253 unreachable("bad polygon mode");
254 }
255 }
256
257 struct bcolor_entry {
258 uint32_t fp32[4];
259 uint64_t ui16;
260 uint64_t si16;
261 uint64_t fp16;
262 uint16_t rgb565;
263 uint16_t rgb5a1;
264 uint16_t rgba4;
265 uint8_t __pad0[2];
266 uint32_t ui8;
267 uint32_t si8;
268 uint32_t rgb10a2;
269 uint32_t z24; /* also s8? */
270 uint64_t srgb;
271 uint8_t __pad1[56];
272 } __attribute__((aligned(128)));
273
274 /* vulkan does not want clamping of integer clear values, differs from u_format
275 * see spec for VkClearColorValue
276 */
277 static inline void
pack_int8(uint32_t * dst,const uint32_t * val)278 pack_int8(uint32_t *dst, const uint32_t *val)
279 {
280 *dst = (val[0] & 0xff) |
281 (val[1] & 0xff) << 8 |
282 (val[2] & 0xff) << 16 |
283 (val[3] & 0xff) << 24;
284 }
285
286 static inline void
pack_int10_2(uint32_t * dst,const uint32_t * val)287 pack_int10_2(uint32_t *dst, const uint32_t *val)
288 {
289 *dst = (val[0] & 0x3ff) |
290 (val[1] & 0x3ff) << 10 |
291 (val[2] & 0x3ff) << 20 |
292 (val[3] & 0x3) << 30;
293 }
294
295 static inline void
pack_int16(uint32_t * dst,const uint32_t * val)296 pack_int16(uint32_t *dst, const uint32_t *val)
297 {
298 dst[0] = (val[0] & 0xffff) |
299 (val[1] & 0xffff) << 16;
300 dst[1] = (val[2] & 0xffff) |
301 (val[3] & 0xffff) << 16;
302 }
303
304 static inline void
tu6_pack_border_color(struct bcolor_entry * bcolor,const VkClearColorValue * val,bool is_int)305 tu6_pack_border_color(struct bcolor_entry *bcolor, const VkClearColorValue *val, bool is_int)
306 {
307 memcpy(bcolor->fp32, val, 4 * sizeof(float));
308 if (is_int) {
309 pack_int16((uint32_t*) &bcolor->fp16, val->uint32);
310 return;
311 }
312 #define PACK_F(x, type) util_format_##type##_pack_rgba_float \
313 ( (uint8_t*) (&bcolor->x), 0, val->float32, 0, 1, 1)
314 PACK_F(ui16, r16g16b16a16_unorm);
315 PACK_F(si16, r16g16b16a16_snorm);
316 PACK_F(fp16, r16g16b16a16_float);
317 PACK_F(rgb565, r5g6b5_unorm);
318 PACK_F(rgb5a1, r5g5b5a1_unorm);
319 PACK_F(rgba4, r4g4b4a4_unorm);
320 PACK_F(ui8, r8g8b8a8_unorm);
321 PACK_F(si8, r8g8b8a8_snorm);
322 PACK_F(rgb10a2, r10g10b10a2_unorm);
323 util_format_x8z24_unorm_pack_z_float((uint8_t*) &bcolor->z24,
324 0, val->float32, 0, 1, 1);
325 PACK_F(srgb, r16g16b16a16_float); /* TODO: clamp? */
326 #undef PACK_F
327 }
328
329 #endif /* TU_UTIL_H */
330