vc4_qir.h revision f87c7008958cdb095efa1cfb29ca8f3c9b9066e4 
1/*
2 * Copyright © 2014 Broadcom
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#ifndef VC4_QIR_H
25#define VC4_QIR_H
26
27#include <stdio.h>
28#include <stdlib.h>
29#include <stdbool.h>
30#include <stdint.h>
31#include <string.h>
32
33#include "util/u_simple_list.h"
34#include "tgsi/tgsi_parse.h"
35
36enum qfile {
37        QFILE_NULL,
38        QFILE_TEMP,
39        QFILE_VARY,
40        QFILE_UNIF,
41};
42
43struct qreg {
44        enum qfile file;
45        uint32_t index;
46};
47
48enum qop {
49        QOP_UNDEF,
50        QOP_MOV,
51        QOP_FADD,
52        QOP_FSUB,
53        QOP_FMUL,
54        QOP_MUL24,
55        QOP_FMIN,
56        QOP_FMAX,
57        QOP_FMINABS,
58        QOP_FMAXABS,
59        QOP_ADD,
60        QOP_SUB,
61        QOP_SHL,
62        QOP_SHR,
63        QOP_ASR,
64        QOP_MIN,
65        QOP_MAX,
66        QOP_AND,
67        QOP_OR,
68        QOP_XOR,
69        QOP_NOT,
70
71        /* Sets the flag register according to src. */
72        QOP_SF,
73
74        /* Note: Orderings of these compares must be the same as in
75         * qpu_defines.h.  Selects the src[0] if the ns flag bit is set,
76         * otherwise 0. */
77        QOP_SEL_X_0_ZS,
78        QOP_SEL_X_0_ZC,
79        QOP_SEL_X_0_NS,
80        QOP_SEL_X_0_NC,
81        /* Selects the src[0] if the ns flag bit is set, otherwise src[1]. */
82        QOP_SEL_X_Y_ZS,
83        QOP_SEL_X_Y_ZC,
84        QOP_SEL_X_Y_NS,
85        QOP_SEL_X_Y_NC,
86
87        QOP_FTOI,
88        QOP_ITOF,
89        QOP_RCP,
90        QOP_RSQ,
91        QOP_EXP2,
92        QOP_LOG2,
93        QOP_VW_SETUP,
94        QOP_VR_SETUP,
95        QOP_PACK_SCALED,
96        QOP_PACK_COLORS,
97        QOP_VPM_WRITE,
98        QOP_VPM_READ,
99        QOP_TLB_DISCARD_SETUP,
100        QOP_TLB_STENCIL_SETUP,
101        QOP_TLB_Z_WRITE,
102        QOP_TLB_COLOR_WRITE,
103        QOP_TLB_COLOR_READ,
104        QOP_VARY_ADD_C,
105
106        QOP_FRAG_X,
107        QOP_FRAG_Y,
108        QOP_FRAG_Z,
109        QOP_FRAG_W,
110        QOP_FRAG_REV_FLAG,
111
112        QOP_UNPACK_8A,
113        QOP_UNPACK_8B,
114        QOP_UNPACK_8C,
115        QOP_UNPACK_8D,
116
117        /** Texture x coordinate parameter write */
118        QOP_TEX_S,
119        /** Texture y coordinate parameter write */
120        QOP_TEX_T,
121        /** Texture border color parameter or cube map z coordinate write */
122        QOP_TEX_R,
123        /** Texture LOD bias parameter write */
124        QOP_TEX_B,
125
126        /**
127         * Texture-unit 4-byte read with address provided direct in S
128         * cooordinate.
129         *
130         * The first operand is the offset from the start of the UBO, and the
131         * second is the uniform that has the UBO's base pointer.
132         */
133        QOP_TEX_DIRECT,
134
135        /**
136         * Signal of texture read being necessary and then reading r4 into
137         * the destination
138         */
139        QOP_TEX_RESULT,
140        QOP_R4_UNPACK_A,
141        QOP_R4_UNPACK_B,
142        QOP_R4_UNPACK_C,
143        QOP_R4_UNPACK_D
144};
145
146struct simple_node {
147        struct simple_node *next;
148        struct simple_node *prev;
149};
150
151struct qinst {
152        struct simple_node link;
153
154        enum qop op;
155        struct qreg dst;
156        struct qreg *src;
157};
158
159enum qstage {
160        /**
161         * Coordinate shader, runs during binning, before the VS, and just
162         * outputs position.
163         */
164        QSTAGE_COORD,
165        QSTAGE_VERT,
166        QSTAGE_FRAG,
167};
168
169enum quniform_contents {
170        /**
171         * Indicates that a constant 32-bit value is copied from the program's
172         * uniform contents.
173         */
174        QUNIFORM_CONSTANT,
175        /**
176         * Indicates that the program's uniform contents are used as an index
177         * into the GL uniform storage.
178         */
179        QUNIFORM_UNIFORM,
180
181        /** @{
182         * Scaling factors from clip coordinates to relative to the viewport
183         * center.
184         *
185         * This is used by the coordinate and vertex shaders to produce the
186         * 32-bit entry consisting of 2 16-bit fields with 12.4 signed fixed
187         * point offsets from the viewport ccenter.
188         */
189        QUNIFORM_VIEWPORT_X_SCALE,
190        QUNIFORM_VIEWPORT_Y_SCALE,
191        /** @} */
192
193        QUNIFORM_VIEWPORT_Z_OFFSET,
194        QUNIFORM_VIEWPORT_Z_SCALE,
195
196        QUNIFORM_USER_CLIP_PLANE,
197
198        /**
199         * A reference to a texture config parameter 0 uniform.
200         *
201         * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
202         * defines texture type, miplevels, and such.  It will be found as a
203         * parameter to the first QOP_TEX_[STRB] instruction in a sequence.
204         */
205        QUNIFORM_TEXTURE_CONFIG_P0,
206
207        /**
208         * A reference to a texture config parameter 1 uniform.
209         *
210         * This is a uniform implicitly loaded with a QPU_W_TMU* write, which
211         * defines texture width, height, filters, and wrap modes.  It will be
212         * found as a parameter to the second QOP_TEX_[STRB] instruction in a
213         * sequence.
214         */
215        QUNIFORM_TEXTURE_CONFIG_P1,
216
217        /** A reference to a texture config parameter 2 cubemap stride uniform */
218        QUNIFORM_TEXTURE_CONFIG_P2,
219
220        QUNIFORM_UBO_ADDR,
221
222        QUNIFORM_TEXRECT_SCALE_X,
223        QUNIFORM_TEXRECT_SCALE_Y,
224
225        QUNIFORM_TEXTURE_BORDER_COLOR,
226
227        QUNIFORM_BLEND_CONST_COLOR,
228        QUNIFORM_STENCIL,
229
230        QUNIFORM_ALPHA_REF,
231};
232
233struct vc4_varying_semantic {
234        uint8_t semantic;
235        uint8_t index;
236        uint8_t swizzle;
237};
238
239struct vc4_compiler_ubo_range {
240        /**
241         * offset in bytes from the start of the ubo where this range is
242         * uploaded.
243         *
244         * Only set once used is set.
245         */
246        uint32_t dst_offset;
247
248        /**
249         * offset in bytes from the start of the gallium uniforms where the
250         * data comes from.
251         */
252        uint32_t src_offset;
253
254        /** size in bytes of this ubo range */
255        uint32_t size;
256
257        /**
258         * Set if this range is used by the shader for indirect uniforms
259         * access.
260         */
261        bool used;
262};
263
264struct vc4_compile {
265        struct vc4_context *vc4;
266        struct tgsi_parse_context parser;
267        struct qreg *temps;
268        /**
269         * Inputs to the shader, arranged by TGSI declaration order.
270         *
271         * Not all fragment shader QFILE_VARY reads are present in this array.
272         */
273        struct qreg *inputs;
274        struct qreg *outputs;
275        struct qreg *consts;
276        struct qreg addr[4]; /* TGSI ARL destination. */
277        uint32_t temps_array_size;
278        uint32_t inputs_array_size;
279        uint32_t outputs_array_size;
280        uint32_t uniforms_array_size;
281        uint32_t consts_array_size;
282        uint32_t num_consts;
283
284        struct vc4_compiler_ubo_range *ubo_ranges;
285        uint32_t ubo_ranges_array_size;
286        uint32_t num_ubo_ranges;
287        uint32_t next_ubo_dst_offset;
288
289        struct qreg line_x, point_x, point_y;
290        struct qreg discard;
291
292        /**
293         * Array of the TGSI semantics of all FS QFILE_VARY reads.
294         *
295         * This includes those that aren't part of the VPM varyings, like
296         * point/line coordinates.
297         */
298        struct vc4_varying_semantic *input_semantics;
299        uint32_t num_input_semantics;
300        uint32_t input_semantics_array_size;
301
302        /**
303         * An entry per outputs[] in the VS indicating what the semantic of
304         * the output is.  Used to emit from the VS in the order that the FS
305         * needs.
306         */
307        struct vc4_varying_semantic *output_semantics;
308
309        struct pipe_shader_state *shader_state;
310        struct vc4_key *key;
311        struct vc4_fs_key *fs_key;
312        struct vc4_vs_key *vs_key;
313
314        uint32_t *uniform_data;
315        enum quniform_contents *uniform_contents;
316        uint32_t uniform_array_size;
317        uint32_t num_uniforms;
318        uint32_t num_outputs;
319        uint32_t num_texture_samples;
320        uint32_t output_position_index;
321        uint32_t output_clipvertex_index;
322        uint32_t output_color_index;
323        uint32_t output_point_size_index;
324
325        struct qreg undef;
326        enum qstage stage;
327        uint32_t num_temps;
328        struct simple_node instructions;
329        uint32_t immediates[1024];
330
331        struct simple_node qpu_inst_list;
332        uint64_t *qpu_insts;
333        uint32_t qpu_inst_count;
334        uint32_t qpu_inst_size;
335        uint32_t num_inputs;
336
337        uint32_t program_id;
338        uint32_t variant_id;
339};
340
341struct vc4_compile *qir_compile_init(void);
342void qir_compile_destroy(struct vc4_compile *c);
343struct qinst *qir_inst(enum qop op, struct qreg dst,
344                       struct qreg src0, struct qreg src1);
345struct qinst *qir_inst4(enum qop op, struct qreg dst,
346                        struct qreg a,
347                        struct qreg b,
348                        struct qreg c,
349                        struct qreg d);
350void qir_remove_instruction(struct qinst *qinst);
351void qir_reorder_uniforms(struct vc4_compile *c);
352void qir_emit(struct vc4_compile *c, struct qinst *inst);
353struct qreg qir_get_temp(struct vc4_compile *c);
354int qir_get_op_nsrc(enum qop qop);
355bool qir_reg_equals(struct qreg a, struct qreg b);
356bool qir_has_side_effects(struct vc4_compile *c, struct qinst *inst);
357bool qir_depends_on_flags(struct qinst *inst);
358bool qir_writes_r4(struct qinst *inst);
359bool qir_reads_r4(struct qinst *inst);
360
361void qir_dump(struct vc4_compile *c);
362void qir_dump_inst(struct vc4_compile *c, struct qinst *inst);
363const char *qir_get_stage_name(enum qstage stage);
364
365void qir_optimize(struct vc4_compile *c);
366bool qir_opt_algebraic(struct vc4_compile *c);
367bool qir_opt_copy_propagation(struct vc4_compile *c);
368bool qir_opt_cse(struct vc4_compile *c);
369bool qir_opt_dead_code(struct vc4_compile *c);
370
371#define QIR_ALU0(name)                                                   \
372static inline struct qreg                                                \
373qir_##name(struct vc4_compile *c)                                        \
374{                                                                        \
375        struct qreg t = qir_get_temp(c);                                 \
376        qir_emit(c, qir_inst(QOP_##name, t, c->undef, c->undef));        \
377        return t;                                                        \
378}
379
380#define QIR_ALU1(name)                                                   \
381static inline struct qreg                                                \
382qir_##name(struct vc4_compile *c, struct qreg a)                         \
383{                                                                        \
384        struct qreg t = qir_get_temp(c);                                 \
385        qir_emit(c, qir_inst(QOP_##name, t, a, c->undef));               \
386        return t;                                                        \
387}
388
389#define QIR_ALU2(name)                                                   \
390static inline struct qreg                                                \
391qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)          \
392{                                                                        \
393        struct qreg t = qir_get_temp(c);                                 \
394        qir_emit(c, qir_inst(QOP_##name, t, a, b));                      \
395        return t;                                                        \
396}
397
398#define QIR_NODST_1(name)                                               \
399static inline void                                                      \
400qir_##name(struct vc4_compile *c, struct qreg a)                        \
401{                                                                       \
402        qir_emit(c, qir_inst(QOP_##name, c->undef, a, c->undef));       \
403}
404
405#define QIR_NODST_2(name)                                               \
406static inline void                                                      \
407qir_##name(struct vc4_compile *c, struct qreg a, struct qreg b)         \
408{                                                                       \
409        qir_emit(c, qir_inst(QOP_##name, c->undef, a, b));       \
410}
411
412QIR_ALU1(MOV)
413QIR_ALU2(FADD)
414QIR_ALU2(FSUB)
415QIR_ALU2(FMUL)
416QIR_ALU2(MUL24)
417QIR_NODST_1(SF)
418QIR_ALU1(SEL_X_0_ZS)
419QIR_ALU1(SEL_X_0_ZC)
420QIR_ALU1(SEL_X_0_NS)
421QIR_ALU1(SEL_X_0_NC)
422QIR_ALU2(SEL_X_Y_ZS)
423QIR_ALU2(SEL_X_Y_ZC)
424QIR_ALU2(SEL_X_Y_NS)
425QIR_ALU2(SEL_X_Y_NC)
426QIR_ALU2(FMIN)
427QIR_ALU2(FMAX)
428QIR_ALU2(FMINABS)
429QIR_ALU2(FMAXABS)
430QIR_ALU1(FTOI)
431QIR_ALU1(ITOF)
432
433QIR_ALU2(ADD)
434QIR_ALU2(SUB)
435QIR_ALU2(SHL)
436QIR_ALU2(SHR)
437QIR_ALU2(ASR)
438QIR_ALU2(MIN)
439QIR_ALU2(MAX)
440QIR_ALU2(AND)
441QIR_ALU2(OR)
442QIR_ALU2(XOR)
443QIR_ALU1(NOT)
444
445QIR_ALU1(RCP)
446QIR_ALU1(RSQ)
447QIR_ALU1(EXP2)
448QIR_ALU1(LOG2)
449QIR_ALU2(PACK_SCALED)
450QIR_ALU1(VARY_ADD_C)
451QIR_NODST_1(VPM_WRITE)
452QIR_NODST_2(TEX_S)
453QIR_NODST_2(TEX_T)
454QIR_NODST_2(TEX_R)
455QIR_NODST_2(TEX_B)
456QIR_NODST_2(TEX_DIRECT)
457QIR_ALU0(FRAG_X)
458QIR_ALU0(FRAG_Y)
459QIR_ALU0(FRAG_Z)
460QIR_ALU0(FRAG_W)
461QIR_ALU0(FRAG_REV_FLAG)
462QIR_ALU0(TEX_RESULT)
463QIR_ALU0(TLB_COLOR_READ)
464QIR_NODST_1(TLB_Z_WRITE)
465QIR_NODST_1(TLB_DISCARD_SETUP)
466QIR_NODST_1(TLB_STENCIL_SETUP)
467
468static inline struct qreg
469qir_R4_UNPACK(struct vc4_compile *c, struct qreg r4, int i)
470{
471        struct qreg t = qir_get_temp(c);
472        qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, r4, c->undef));
473        return t;
474}
475
476static inline struct qreg
477qir_SEL_X_0_COND(struct vc4_compile *c, int i)
478{
479        struct qreg t = qir_get_temp(c);
480        qir_emit(c, qir_inst(QOP_R4_UNPACK_A + i, t, c->undef, c->undef));
481        return t;
482}
483
484static inline struct qreg
485qir_UNPACK_8(struct vc4_compile *c, struct qreg src, int i)
486{
487        struct qreg t = qir_get_temp(c);
488        qir_emit(c, qir_inst(QOP_UNPACK_8A + i, t, src, c->undef));
489        return t;
490}
491
492static inline struct qreg
493qir_POW(struct vc4_compile *c, struct qreg x, struct qreg y)
494{
495        return qir_EXP2(c, qir_FMUL(c,
496                                    y,
497                                    qir_LOG2(c, x)));
498}
499
500#endif /* VC4_QIR_H */
501