1 /*
2  * Copyright 2011 Christoph Bumiller
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 shall be included in
12  * all copies or substantial portions of the Software.
13  *
14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20  * OTHER DEALINGS IN THE SOFTWARE.
21  */
22 
23 #ifndef __NV50_IR_DRIVER_H__
24 #define __NV50_IR_DRIVER_H__
25 
26 #include "pipe/p_shader_tokens.h"
27 
28 #include "util/blob.h"
29 #include "tgsi/tgsi_util.h"
30 #include "tgsi/tgsi_parse.h"
31 #include "tgsi/tgsi_scan.h"
32 
33 struct nir_shader_compiler_options;
34 
35 /*
36  * This struct constitutes linkage information in TGSI terminology.
37  *
38  * It is created by the code generator and handed to the pipe driver
39  * for input/output slot assignment.
40  */
41 struct nv50_ir_varying
42 {
43    uint8_t slot[4]; /* native slots for xyzw (addresses in 32-bit words) */
44 
45    unsigned mask     : 4; /* vec4 mask */
46    unsigned linear   : 1; /* linearly interpolated if true (and not flat) */
47    unsigned flat     : 1;
48    unsigned sc       : 1; /* special colour interpolation mode (SHADE_MODEL) */
49    unsigned centroid : 1;
50    unsigned patch    : 1; /* patch constant value */
51    unsigned regular  : 1; /* driver-specific meaning (e.g. input in sreg) */
52    unsigned input    : 1; /* indicates direction of system values */
53    unsigned oread    : 1; /* true if output is read from parallel TCP */
54 
55    ubyte id; /* TGSI register index */
56    ubyte sn; /* TGSI semantic name */
57    ubyte si; /* TGSI semantic index */
58 };
59 
60 #ifndef NDEBUG
61 # define NV50_IR_DEBUG_BASIC     (1 << 0)
62 # define NV50_IR_DEBUG_VERBOSE   (2 << 0)
63 # define NV50_IR_DEBUG_REG_ALLOC (1 << 2)
64 #else
65 # define NV50_IR_DEBUG_BASIC     0
66 # define NV50_IR_DEBUG_VERBOSE   0
67 # define NV50_IR_DEBUG_REG_ALLOC 0
68 #endif
69 
70 struct nv50_ir_prog_symbol
71 {
72    uint32_t label;
73    uint32_t offset;
74 };
75 
76 #define NVISA_GF100_CHIPSET    0xc0
77 #define NVISA_GK104_CHIPSET    0xe0
78 #define NVISA_GK20A_CHIPSET    0xea
79 #define NVISA_GM107_CHIPSET    0x110
80 #define NVISA_GM200_CHIPSET    0x120
81 #define NVISA_GV100_CHIPSET    0x140
82 
83 struct nv50_ir_prog_info_out;
84 
85 /* used for the input data and assignSlot interface */
86 struct nv50_ir_prog_info
87 {
88    uint16_t target; /* chipset (0x50, 0x84, 0xc0, ...) */
89 
90    uint8_t type; /* PIPE_SHADER */
91 
92    uint8_t optLevel; /* optimization level (0 to 3) */
93    uint8_t dbgFlags;
94    bool omitLineNum; /* only used for printing the prog when dbgFlags is set */
95 
96    struct {
97       uint32_t smemSize;  /* required shared memory per block */
98       uint8_t sourceRep;  /* PIPE_SHADER_IR_* */
99       const void *source;
100    } bin;
101 
102    union {
103       struct {
104          uint32_t inputOffset; /* base address for user args */
105          uint32_t gridInfoBase;  /* base address for NTID,NCTAID */
106          uint16_t numThreads[3]; /* max number of threads */
107       } cp;
108    } prop;
109 
110    struct {
111       int8_t genUserClip;        /* request user clip planes for ClipVertex */
112       uint8_t auxCBSlot;         /* driver constant buffer slot */
113       uint16_t ucpBase;          /* base address for UCPs */
114       uint16_t drawInfoBase;     /* base address for draw parameters */
115       uint16_t alphaRefBase;     /* base address for alpha test values */
116       int8_t viewportId;         /* output index of ViewportIndex */
117       bool mul_zero_wins;        /* program wants for x*0 = 0 */
118       bool nv50styleSurfaces;    /* generate gX[] access for raw buffers */
119       uint16_t texBindBase;      /* base address for tex handles (nve4) */
120       uint16_t fbtexBindBase;    /* base address for fbtex handle (nve4) */
121       uint16_t suInfoBase;       /* base address for surface info (nve4) */
122       uint16_t bindlessBase;     /* base address for bindless image info (nve4) */
123       uint16_t bufInfoBase;      /* base address for buffer info */
124       uint16_t sampleInfoBase;   /* base address for sample positions */
125       uint8_t msInfoCBSlot;      /* cX[] used for multisample info */
126       uint16_t msInfoBase;       /* base address for multisample info */
127       uint16_t uboInfoBase;      /* base address for compute UBOs (gk104+) */
128    } io;
129 
130    /* driver callback to assign input/output locations */
131    int (*assignSlots)(struct nv50_ir_prog_info_out *);
132 };
133 
134 /* the produced binary with metadata */
135 struct nv50_ir_prog_info_out
136 {
137    uint16_t target; /* chipset (0x50, 0x84, 0xc0, ...) */
138 
139    uint8_t type; /* PIPE_SHADER */
140 
141    struct {
142       int16_t maxGPR;     /* may be -1 if none used */
143       uint32_t tlsSpace;  /* required local memory per thread */
144       uint32_t smemSize;  /* required shared memory per block */
145       uint32_t *code;
146       uint32_t codeSize;
147       uint32_t instructions;
148       void *relocData;
149       void *fixupData;
150    } bin;
151 
152    struct nv50_ir_varying sv[PIPE_MAX_SHADER_INPUTS];
153    struct nv50_ir_varying in[PIPE_MAX_SHADER_INPUTS];
154    struct nv50_ir_varying out[PIPE_MAX_SHADER_OUTPUTS];
155    uint8_t numInputs;
156    uint8_t numOutputs;
157    uint8_t numPatchConstants; /* also included in numInputs/numOutputs */
158    uint8_t numSysVals;
159 
160    union {
161       struct {
162          bool usesDrawParameters;
163       } vp;
164       struct {
165          uint8_t outputPatchSize;
166          uint8_t partitioning;    /* PIPE_TESS_PART */
167          int8_t winding;          /* +1 (clockwise) / -1 (counter-clockwise) */
168          uint8_t domain;          /* PIPE_PRIM_{QUADS,TRIANGLES,LINES} */
169          uint8_t outputPrim;      /* PIPE_PRIM_{TRIANGLES,LINES,POINTS} */
170       } tp;
171       struct {
172          uint8_t outputPrim;
173          unsigned instanceCount;
174          unsigned maxVertices;
175       } gp;
176       struct {
177          unsigned numColourResults;
178          bool writesDepth           : 1;
179          bool earlyFragTests        : 1;
180          bool postDepthCoverage     : 1;
181          bool usesDiscard           : 1;
182          bool usesSampleMaskIn      : 1;
183          bool readsFramebuffer      : 1;
184          bool readsSampleLocations  : 1;
185          bool separateFragData      : 1;
186       } fp;
187    } prop;
188 
189    struct {
190       uint8_t clipDistances;     /* number of clip distance outputs */
191       uint8_t cullDistances;     /* number of cull distance outputs */
192       int8_t genUserClip;        /* request user clip planes for ClipVertex */
193       uint8_t instanceId;        /* system value index of InstanceID */
194       uint8_t vertexId;          /* system value index of VertexID */
195       uint8_t edgeFlagIn;
196       uint8_t edgeFlagOut;
197       uint8_t fragDepth;         /* output index of FragDepth */
198       uint8_t sampleMask;        /* output index of SampleMask */
199       uint8_t globalAccess;      /* 1 for read, 2 for wr, 3 for rw */
200       bool fp64;                 /* program uses fp64 math */
201       bool layer_viewport_relative;
202    } io;
203 
204    uint8_t numBarriers;
205 
206    void *driverPriv;
207 };
208 
209 #ifdef __cplusplus
210 extern "C" {
211 #endif
212 
213 const struct nir_shader_compiler_options *
214 nv50_ir_nir_shader_compiler_options(int chipset);
215 
216 extern int nv50_ir_generate_code(struct nv50_ir_prog_info *,
217                                  struct nv50_ir_prog_info_out *);
218 
219 extern void nv50_ir_relocate_code(void *relocData, uint32_t *code,
220                                   uint32_t codePos,
221                                   uint32_t libPos,
222                                   uint32_t dataPos);
223 
224 extern void
225 nv50_ir_apply_fixups(void *fixupData, uint32_t *code,
226                      bool force_per_sample, bool flatshade,
227                      uint8_t alphatest);
228 
229 /* obtain code that will be shared among programs */
230 extern void nv50_ir_get_target_library(uint32_t chipset,
231                                        const uint32_t **code, uint32_t *size);
232 
233 
234 #ifdef __cplusplus
235 namespace nv50_ir
236 {
237    class FixupEntry;
238    class FixupData;
239 
240    void
241    gk110_interpApply(const nv50_ir::FixupEntry *entry, uint32_t *code,
242                      const nv50_ir::FixupData& data);
243    void
244    gm107_interpApply(const nv50_ir::FixupEntry *entry, uint32_t *code,
245                      const nv50_ir::FixupData& data);
246    void
247    nv50_interpApply(const nv50_ir::FixupEntry *entry, uint32_t *code,
248                     const nv50_ir::FixupData& data);
249    void
250    nvc0_interpApply(const nv50_ir::FixupEntry *entry, uint32_t *code,
251                     const nv50_ir::FixupData& data);
252    void
253    gv100_interpApply(const nv50_ir::FixupEntry *entry, uint32_t *code,
254                      const nv50_ir::FixupData& data);
255    void
256    gk110_selpFlip(const nv50_ir::FixupEntry *entry, uint32_t *code,
257                   const nv50_ir::FixupData& data);
258    void
259    gm107_selpFlip(const nv50_ir::FixupEntry *entry, uint32_t *code,
260                   const nv50_ir::FixupData& data);
261    void
262    nvc0_selpFlip(const nv50_ir::FixupEntry *entry, uint32_t *code,
263                  const nv50_ir::FixupData& data);
264    void
265    gv100_selpFlip(const nv50_ir::FixupEntry *entry, uint32_t *code,
266                   const nv50_ir::FixupData& data);
267 }
268 #endif
269 
270 extern void
271 nv50_ir_prog_info_out_print(struct nv50_ir_prog_info_out *);
272 
273 /* Serialize a nv50_ir_prog_info structure and save it into blob */
274 extern bool
275 nv50_ir_prog_info_serialize(struct blob *, struct nv50_ir_prog_info *);
276 
277 /* Serialize a nv50_ir_prog_info_out structure and save it into blob */
278 extern bool MUST_CHECK
279 nv50_ir_prog_info_out_serialize(struct blob *, struct nv50_ir_prog_info_out *);
280 
281 /* Deserialize from data and save into a nv50_ir_prog_info_out structure
282  * using a pointer. Size is a total size of the serialized data.
283  * Offset points to where info_out in data is located. */
284 extern bool MUST_CHECK
285 nv50_ir_prog_info_out_deserialize(void *data, size_t size, size_t offset,
286                                   struct nv50_ir_prog_info_out *);
287 
288 #ifdef __cplusplus
289 }
290 #endif
291 
292 #endif // __NV50_IR_DRIVER_H__
293